Fix(MInference): fix the residual

[iofu728]

What does this PR do?

• Fixes the residual;
• Fixed self._supports_num_logits_to_keep;

Before submitting

• This PR fixes a typo or improves the docs (you can dismiss the other checks if that's the case).
• Was this discussed/approved via a Github issue? Please add a link
  to it if that's the case.
• Did you make sure to update the documentation with your changes?
• Did you write any new necessary tests?

Who can review?

@iofu728

[Copilot]

Pull Request Overview

This PR fixes residual computation in the Llama decoder layer and improves the handling of logits support, while updating the constructor signature for DynamicCacheWithRepeat.

• Uses hidden_states.clone() to create an independent residual copy in the decoder layer.
• Adds additional checks for supported logits properties in cache preparation.
• Updates the init signature in DynamicCacheWithRepeat to require a config parameter.

Reviewed Changes

Copilot reviewed 2 out of 2 changed files in this pull request and generated no comments.

File	Description
minference/patch.py	Clones hidden states in the Llama decoder layer to avoid unintended mutations.
minference/modules/kvcompression.py	Adds conditional checks for logits support and updates DynamicCacheWithRepeat's init.

Comments suppressed due to low confidence (1)

minference/modules/kvcompression.py:370

• Changing the init signature to require a 'config' argument may break backward compatibility if existing instantiations do not pass this parameter. Verify that all usages of DynamicCacheWithRepeat are updated accordingly.

def __init__(self, config, *args, **kwargs):