fix: add gguf wrapper in quantized gemma3

[junjunjd]

Add Gguf wrapper in quantized Gemma3 loader to fix missing metadata lookups.

Fixes #3215.

[ivarflakstad]

@junjunjd Thanks for this!
I tried running the embedding-gemma model mentioned in the issue and unfortunately this doesn't fix the issue. Luckily the problem isn't has complex as it may seem.
Default gemma3 uses the prefix gemma3.*, while the embedding-gemma model uses gemma-embedding.* (ref).
ModelWeights in quantized_gemma3.rs is hard coded to use gemma3.
I'll post the findings in the issue as well.