Add ignore_accents option to by_name()

[gefleury]

Adds accent-insensitive matching to by_name(), a follow-up to the case_sensitive and match options from #100.

• New ignore_accents parameter (default False, so existing behaviour is unchanged).
• When True, accents and other diacritical marks are stripped before matching (e.g. "Republique francaise" matches "République française"). Combine with case_sensitive=False to absorb case differences as well.
• Uses NFD (accents only), not NFKD, so Unicode compatibility folding (e.g. superscripts
  like m³→m3) is deliberately not applied here. That's a separate concern that could
  be handled later if needed.
• The generated modules now import unicodedata (added to the preamble in translator.py
  only for modules that have by_name).
• One test added in test_regressions.py

[apdavison]

This doesn't handle Ł ø đ æ ß ə ı

For example:

>>> import unicodedata
>>> s = "Łódź"
>>> unicodedata.normalize("NFD", s)
'Łódź'

[gefleury]

You're right, I missed it.

• "NFKD" (instead of "NFD") doesn't help either
• unidecode handles those letters

from unidecode import unidecode

for s in ["Ł", "ø", "đ", "æ", "ß", "ə", "ı", "µA", "m³", "Ελλάδα"]:
    print(f"{s} -> {unidecode(s)}")

### Output
Ł -> L
ø -> o
đ -> d
æ -> ae
ß -> ss
ə -> @
ı -> i
µA -> uA
m³ -> m3
Ελλάδα -> Ellada

but it adds a runtime dependency and transliterates too aggressively (e.g. it maps ə to @, which is wrong for eg Azerbaijan's synonym "Azərbaycan").

• My preferred option: keep NFD for accents and add a small explicit map (no dependency, full control):

_SPECIAL = str.maketrans({
    "Ł": "L", "ł": "l",
    "Ø": "O", "ø": "o",
    "Đ": "D", "đ": "d",
    "Æ": "AE", "æ": "ae",
    "ß": "ss",
    "Ə": "E", "ə": "e",
    "ı": "i",
})

def remove_accents(s):
    s = s.translate(_SPECIAL)                 
    nfd_form = unicodedata.normalize("NFD", s)
    return "".join(c for c in nfd_fo

Happy to implement it this way if you agree

[apdavison]

yes, I'm happy with that approach

[gefleury]

OK. It's now implemented in 21e0397 with an updated test.

[apdavison]

Since unicodedata is used in only one specific place in the code, it might be better to import unicodedata inside the remove_accents() function.

[gefleury]

Yes, that's simpler. Done in 3eed40f