[Feature Request] Homogenize language codes across model families

[FerLuisxd]

Currently, RunOptions.language validation differs across model families:

• SenseVoice: Requires 2-letter ISO codes (ja, zh, yue, ko, en). Passing ja-JP fails with status 10 (unsupported language).
• Nemotron 3.5 / Whisper: Requires 5-letter locale codes (ja-JP, en-US, es-ES). Passing ja fails with status 10.

Proposed Solution

Add internal aliases inside transcribe.cpp so that both 2-letter codes (ja, zh, es) and 5-letter locale codes (ja-JP, zh-CN, es-ES) map transparently for any model family without throwing unsupported language (status 10).
Or a way so after loading the model it should also return the available supported languages?

[cjpais]

Yes, this probably needs to be done in some ways