It seems I misused the term “convention”. I now see that I cannot use it in the context of a constructed language as one does in the context of natural languages.

I used it in the latter context (which was wrong). In natural languages “convention” means anything that is arbitrarily chosen (whichout any rationale) and generally agreed upon.
The fact that in English the sound /b/ is represented by the grapheme B and not by A, X, 8 or ß is a convention. There is no logic to it, that /b/ should be represented by a stick with two loops.
The fact that in English the sound /dog/ refers to a barking mammal is convention. It could well have been any other sound.
The fact that in English you say “I am fine” and not “Fine I am” is convention.
Almost everything is a natural language is by convention.

Rules in natural languages do not define the language. The rules describe the language. (Descriptive grammar)

And you are completely right: In constructed languages the rules indeed define the language. (Prescriptive grammar).

In English there is no rule that states that dog means braking animal; we just all agreed upon it and use it that way. In Esperanto there is indeed a rule (in this case an item in the UV) that defines that hundo refers to such an animal.

The problem is that Esperanto is not fully defined by the 16 rules in the Fundamenta Gramatiko. That’s why there are works like PMEG with hundreds of rules. In the original 16 rules, we cannot find anything that bars the sentence “mi Parizas hunde ĉi ĉi”. That’s the point where semantics and convention kick in.

Unfortunately you cannot write a grammar without taking into consideration semantics and pragmatics. And for me personally, that's a big "unfortunately". It took me years and decades to accept that fact. I used to think, that a grammar was a beautiful, mathematical and logical palace with white towers and an impressive moat, untarnished by ugly and squishy things like semantics and pragmatics. Alas, I was wrong.

Nonetheless, I am still convinced that semantics and pragmatics are the idiot brothers of grammar.

The following might be an argument against "-in" instead of "-ul-in" / "-ist-in" / "-ant-in" etc:

In PIV we find:
bovino. Ino de bovo.
bovinisto. Tiu, kiu prizorgas la bovinojn.
bovisto. Tiu, kiu prizorgas la bovojn.

Bovo is a head of cattle.
Bovino is a cow.
Bovisto is a cattle farmer (cattleherd), both the cattle and the farmer can be of any sex.
Bovinisto is a cow farmer (cowherd), the farmer can be of any sex, the cattle are female.
Bovistino is a female cattle farmer (cattleherdess), the farmer is female, the cattle of any sex.
Bovinistino is a female cow farmer (cowherdess), the farmer is female and the cattle are, too.

If you parse bovino as bov-(ino) as you did with knabec-(ino), then bovino and bovistino would mean the same thing.

If you parse bovino as bov-(ino) as you did with knabec-(ino), then bovino and bovistino would mean the same thing.

I feel kind of slack-jawed saying this, but I am having trouble following that. I've read it a bunch of times, and I'm sure it's my own blindness, but I don't come to that conclusion. For the farmer, the -ist part is necessary. In my mind, bovino cannot mean a farmer as it lacks the bit (-ist) that contributes to the farmer meaning; it can only mean a cow.

Unless we take the position that -ino always means human the way -ulo does. So bovino would be a bova ino, which may not be a compliment. That just outright doesn't work, though, as it makes it impossible to refer to real cows, and is not the position I'm taking. What we are really referring to with bovino is an ina bovo (but with a stronger binding than the two words apart, so it's not a feminine bovo but a fully female one), and we combine the in with bovo by putting it at the end instead of at the beginning because it's traditionally (and explicitly) a suffix.

I think there is a "what makes sense" distinction to be made, but that's not a foreign situation in Esperanto. When something can be interpreted two ways, one of which makes no sense, then the way that makes sense is the correct interpretation (although the one that makes no sense is thus also ripe for incorporation into jokes involving the way that does make sense).

All that aside, I'm still not claiming that I am correct, only that this is not the way I've been thinking about -in, thinking which may be fuzzier than it should be, and I need to think harder about it.

I think part of this fuzziness stems from the fact that at some point in the past Esperantists started using “words” which Zamenhof used exclusively as derivational suffixes, as standalone semantic words. These words include in, ig, iĝ, eg, etc.

Zamenhovian fariĝi became iĝi, Zamenhovian devigi became igi, Zamenhovian grandega became ega.

Let’s look at the lemma for “o” and “in” in the UV.

UV:in' marque le féminin; ex. patr' père ― patr'in' mère | ending of feminine words; e. g. bov' ox ― bov'in' cow | bezeichnet das weibliche Geschlecht; z. B. patr' Vater ― patr'in' Mutter; fianĉ' Bräutigam ― fianĉ'in' Braut | женскій полъ; напр. patr' отецъ ― patr'in' мать; fianĉ' женихъ ― fianĉ'in' невѣста | oznacza płeć źeńską; np. patr' ojciec ― patr'in' matka; kok' kogut ― kok'in' kura.

UV:o marque le substantif | ending of nouns (substantive) | bezeichnet das Substantiv | означаетъ существительное | oznacza rzeczownik.

Both are marked as endings, however the apostrophe at the end of in’ tells as, that in’ must still be followed by another morpheme.

One might argue that “ino” is not a complete word at all, because per definitionem it is only the ending of feminine words and by using only “ino” we do not know which word has to be feminine.

If we accept that standalone “ino” means “a female human/animal/god” (Where did the human/animal/god part come from?), then shouldn’t we also accept the standalone word “o” meaning “a substantive”?

Virino, bovino, diino, feino are all fine and proper, but “ino” alone, at least from an analytical point of view, is fuzzy.

The morpheme “in” marks a quality of something, the quality of being female. The morpheme “bel” also marks a quality of something, the quality of being beautiful. If “ino” means “a female individual” why doesn’t “belo” mean “a beautiful individual”? If “belo” means “beauty” why doesn’t “ino” mean “femininity” (Is that a word?).

Unfortunately this train of thought opens a whole new can of worms: the idea that roots by itself belong to a part of speech. The broŝo/kombo issue. We do not want to enter these waters.

Another problem is that Z doesn’t make a difference between grammatical gender and biological sex. “Ending of feminine words” apparently doesn’t mean “ending of words belonging to the feminine grammatical gender”, but “ending of words referring to individuals of female sex”. However, as Esperanto has no grammatical gender (even li/ŝi/ĝi depend only on biological sex and do not mark any grammatical gender; also in a language without syntactical agreement, talking about gender is non-sensical), only the latter interpretation is possible.

Another thing is the order of precendence when forming compound words. Barring prefixes (mal/ge/bo/etc) the order is left to right. A vaporŝipestro is a ((vapor’ŝip)’estr)’o, that is the captain of a steamboat; and it is not e.g. a (vapor’(ŝip’estr))’o which would be a steam-captain of a boat (maybe a steam elemental from D&D terrorising the Caribbean). Hence knabecino should be parsed as ((knab’ec)’in)’o which is a substantive referring to female boyishness; and not (knab’ec)’(in’o) as you read it which would be a woman exposing boyishness (tomboy). Now knabeculino works as expected: (((knab’ec)’ul)’in)’o which is a noun (o) referring to a female (in) individual (ul) exposing certain traits (ec) of a boy (knab).

Yes, you do have a point with the "new" words being formed out of affixes. The problem now becomes that this isn't going to go away. It isn't just me that does this, it's now rampant. It's hard to find text where this doesn't appear anymore. But as you say, maybe the affixes don't work like regular roots, or only sort of do. That might be an issue.

Also, yes, knabeculino does what I expected knabecino to do. Before, I thought of the -ul as being redundant, since to my way of thinking the -in implied an -ul. But I believe that this was a nuance I missed. The -ul is necessary.