Llama 4 Scout and Maverick archs

[MartinGleize]

What does this PR do? Please describe:
This PR adds Llama 4 Scout and Maverick support to fairseq2.

Test plan:

• Llama 4 Scout native inference recipe support:

torchrun --standalone --nproc-per-node 8 --no-python \
    fairseq2 lm generate $SAVE_DIR --no-sweep-dir --config \
    model.name=$CKPT \
    tokenizer.name=$TOKENIZER \
    gang.tensor_parallel_size=8 \
    seq_generator.config.temperature=0.0 \
    seq_generator.batch_size=256 \
    dataset.path=$DATASET

• No regression on Llama 3 inference recipe
• Llama 4 Scout parity: verified against llama-stack on gsm8k test. 95% 0-shot acc.
• Llama 4 Scout finetuning: recipe should work but no guarantee on effectiveness (token choice routing likely needs a load balancing loss, this will be another PR).

Does your PR introduce any breaking changes? If yes, please list them:
No

Check list:

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