Fix language codes not recognized correctly

[Aunirbhan]

Problem

get_zim_language_metadata() misses ISO 639-3 codes like gla, nap, and oji because it only looks up 2-letter keys in ISO_MATRIX. They resolve to None, producing an empty language list, and the ZIM crashes.

Solution

• Track unresolved codes and mention them through a warning.
• Raises a clear error pointing to --zim-languages when nothing resolves.
• Added ISO_MATRIX_REV fallback so codes already known as values (like gd→gla, oj→oji) resolve correctly.
• @benoit74 I also added nap to ZIM_LANGUAGES_MAP since you mentioned it on the issue thread.

Before / After

Before — get_zim_language_metadata() on upstream:

get_zim_language_metadata(["gla"], books)  →  []  # silent, no warning
get_zim_language_metadata(["nap"], books)  →  []
get_zim_language_metadata(["oji"], books)  →  []
get_zim_language_metadata(["myn"], books)  →  []

After:

get_zim_language_metadata(["gla"], books)  →  ["gla"]
get_zim_language_metadata(["nap"], books)  →  ["nap"]
get_zim_language_metadata(["oji"], books)  →  ["oji"]

get_zim_language_metadata(["myn"], books)  →  []
# WARNING: "Could not resolve ZIM language metadata for: myn"

get_zim_language_metadata(["myn", "nai"], books)  →  []
# WARNING: "Could not resolve ZIM language metadata for: myn, nai"
# ValueError: "Cannot resolve language metadata for: myn, nai.
#              Use --zim-languages to override."

Passes hatch run test:run

Closes #337

[benoit74]

Thank you. I feel like we are on the right path.

What needs to be enhanced:

• I prefer former logic where we build the language_counts in a single line
• we can probably write the same single line to detect any unresolved code (something like unresolved = [lang for lang in languages if not ZIM_LANGUAGES_MAP.get(lang, [ISO_MATRIX.get(lang, None)])] (to be updated with ISO_MATRIX_REV as well)
• we should fail the scraper immediately when one language is unresolved, not issue a warning, this needs to be fixed rather than silently creating ZIMs with incorrect metadata
• we can probably extract the "complex" logic to get zim_langs from a single language in a dedicated function
• we need to add tests to this
• we need a CHANGELOG entry