Update modeling_mamba2.py, fix pad size (#32599)
* Update modeling_mamba2.py
Fix pad_size calculation to ensure it's less than self.chunk_size
* [run_slow] mamba2
* [run-slow] mamba2
* [run-slow] Add @require_read_token decorator to failing tests for token propagation
* [run_slow] mamba2
