ESMFold2: expose MSA inference-diversity options

[faustomilletari]

What does this PR do?

Port the MSA inference-time diversity feature to the esmfold2 model:

• modeling_esmfold2_common.py: add maybe_subsample_msa and maybe_apply_msa_column_masking helpers (operate on [B, depth, L] encoder inputs, always keep query row 0).
• modeling_esmfold2.py + modeling_esmfold2_experimental.py: new forward kwargs msa_max_depth (1024) / msa_column_mask_rate (0.1) / msa_subsample_at_inference (True). Column masking is applied once before the recycling loop (shared across loops); row subsampling is drawn fresh per loop inside the loop, matching the pro-msa setup.

Defaults are hardcoded in the public signatures (not model config), so these are pure inference-time knobs. The diffusion sampler / num_sampling_steps schedule is untouched.

Fixes #
absence of inference-time diversity knobs in implementation

Before submitting

• This PR fixes a typo or improves the docs (you can dismiss the other checks if that's the case).
• Did you read the contributor guideline,
  Pull Request section?
• Was this discussed/approved via a Github issue or the forum? Please add a link
  to it if that's the case.
• Did you make sure to update the documentation with your changes? Here are the
  documentation guidelines, and
  here are tips on formatting docstrings.
• Did you write any new necessary tests?

Who can review?

Anyone in the community is free to review the PR once the tests have passed. Feel free to tag
members/contributors who may be interested in your PR.