Refactor/understanding prepare inputs padded

[tomasruizt]

I'm trying to understand how prepare_inputs() works.

• prepare_inputs_padded() states that it updates common_attn_metadata. However, the code seems rather to make a copy only.
• I introduced assertions checking that all content is the same, and even make the method return the unmodified common_attn_metadata. Its has no negative impact.
• I run pytest tests/v1/spec_decode/test_eagle.py all tests pass (some are skipped).

=== 29 passed, 20 skipped, 2 warnings in 44.02s ===

• I run pytest tests/v1/e2e/test_spec_decode.py -k test_eagle_correctness -s, all tests pass (some are skipped).

Am I missing something? If not this is an opportunity to simplify the prepare_inputs_padded() method.