[Draft] hg model integration (v0)

[zyaoj]

What does this PR do? Please describe:

• A simple model family implementation to wrap models/processor/tokenizer loaded from transformers auto classes
• Currently support 4 types of Auto classes (auto, causal lm, seq2seq lm, custom)
• Added basic asset examples for qwen3 and qwen2.5 omni (+doc and tests)
• Exposed some additional convenience api-s (free to remove them)
• Wrapping (yet another) layer for hg tokenizers (maybe not ideal but trying to make the interfaces similar to the original huggingface ones to help with the migration)
• [nit] added .coveragerc and some gitignore for dev and test

Loading qwen 2.5 omni should be as simple as:

from fairseq2.models.hg import get_hg_model_hub, get_hg_tokenizer_hub

name = "hg_qwen25_omni_3b"

# Load a pre-configured model
model_hub = get_hg_model_hub()
model = model_hub.load_model(name)

# Load the corresponding tokenizer
tokenizer_hub = get_hg_tokenizer_hub()
tokenizer = tokenizer_hub.load_tokenizer(name)

Here's a paste if you want to check the full log.

Fixes #1301

TBD:

• A wrapper class to handle AutoXXX.from_pretrained(...) -> Any types
• Device placement and customized sharder
• Integration w/ post-training recipes

Does your PR introduce any breaking changes? If yes, please list them:

• Added tests for hg models so it requires installation of [hg] (e.g. `pip install -e '.[hg]') for the full test suite to pass
• For hg tokenizers, added logic to retrieve default special token idx if not overridden by user input

Check list:

• [x] Was the content of this PR discussed and approved via a GitHub issue? (no need for typos or documentation improvements)
• [x] Did you read the contributor guideline?
• [x] Did you make sure that your PR does only one thing instead of bundling different changes together?
• [x] Did you make sure to update the documentation with your changes? (if necessary)
• [x] Did you write any new necessary tests?
• [x] Did you verify new and existing tests pass locally with your changes?
• [ ] Did you update the CHANGELOG? (no need for typos, documentation, or minor internal changes)