Take subject='abbrev' into account again

[percevalw]

Hi ! First, thank you for this great package !

This PR aims at fixing the abbrev mode of subjects when running .conjugate on a verb.

Currently (Python 3.11, mlconjug3==3.11.0), running the first example of the docs:

from mlconjug3 import Conjugator

# initialize the conjugator
conjugator = Conjugator()

# conjugate the verb "parler"
verb = conjugator.conjugate("parler")
# OR verb = conjugator.conjugate("parler", subject="abbrev")

# print all the conjugated forms as a list of tuples.
print(verb.iterate())

print(verb["Indicatif"]["Présent"]["1s"])

raises a KeyError: '1s', because it runs VerbXX._load_conjug twice
(without, store the results in the full_forms attribute, then with abbreviations),
and dismisses the second call, therefore not populating the abbreviated forms of subjects.

This was not the past behavior (at least in 3.9.0), and may cause issues in dependent packages (at least edsnlp)

This can be fixed by assigning the full_forms dict into the conjug_info attribute of verbs before perfoming
the first call.

[SekouDiaoNlp]

Hi @percevalw , thank you for submitting the PR.

I have been pretty busy lately but now have time to dedicate to this project.

i will review your changes shortly and let you know when they are merged.

Thanks again for your feedback.