RAG support for other languages

[supercamel]

One of the coolest things about running local models is that you can fine tune them for niche languages (like I have done for Vala) or subsets of languages (like Arduino). Fine tuning only goes so far, of course. RAG in TabbyML only works for a select few languages, and does not work for these niche languages.

A possible solution would be to configure Tabby to recognize certain file extensions as a particular language. Vala is syntactically very similar to C#, and Arduino language is a subset of C++, so we should be able to configure TabbyML to recognize and parse .vala files as C# source, and .ino as C++. I think that this would be acceptable for RAG purposes.

Long term, it would be nice to be easily able to add tree-sitter implementations for other languages, but until then, configuration to recognize file extensions as a supported language would suffice.

---

Please reply with a 👍 if you want this feature.

[wsxiaoys]

See #553 as an example for adding language support (it's a bit outdated as we refactor all configuration into languages.toml, but can still be used as a reference)