[AutoDeploy][Bug]: Fix the Torch kernel for SSM layers

System Info

The Torch kernel does not handle the mixed batch (one batch with both prefill and decode) for TRT LLM.

Requirement:

1. Fix the torch backend kernels
2. Update the unit test.

Who can help?

No response

Information

• The official example scripts
• My own modified scripts

Tasks

• An officially supported task in the examples folder (such as GLUE/SQuAD, ...)
• My own task or dataset (give details below)

Reproduction

N/A

Expected behavior

The expected output from torch backend should be the same / or approximately same as the triton fast kernels.

actual behavior

The torch backend could only have the prefill-only and decode-only batch.

additional notes

N/A

Before submitting a new issue...

• Make sure you already searched for relevant issues, and checked the documentation and examples for answers to frequently asked questions.