Fix HuBERT apply_mask logic and Python 3.11 compatibility

[SYNX7007]

Before submitting

• Was this discussed/approved via a Github issue? (Fixes apply_mask in HuBERT does not apply expected percentage of masks if the input shapes differ. which is true in the case of fintuning. #5642)
• Did you read the contributor guideline?
• Did you make sure to update the docs? (No doc changes needed)
• Did you write any new necessary tests? (Verified via reproduction script)

What does this PR do?

Fixes #5642.

1. Fix HuBERT Masking Logic

The apply_mask method in fairseq/models/hubert/hubert.py was calling compute_mask_indices with the default require_same_masks=True. This caused variable-length sequences in the same batch to be masked incorrectly (constrained by the shortest sequence). I set require_same_masks=False to resolve this.

2. Python 3.11 Compatibility

Fixed a ValueError in fairseq/dataclass/configs.py caused by mutable default arguments in dataclasses (common: CommonConfig = CommonConfig()), which is not allowed in Python 3.11+. Switched to field(default_factory=...).

Verification

Verified with a reproduction script simulating a batch with lengths [100, 50].

• Before Fix: Length 100 sample was under-masked (~21% instead of 65%).
• After Fix: Length 100 sample is correctly masked (~47-65%).

PR review

Ready for review!