[BugFix][Mamba] Fix PtrOffsetInfo AxisInfo assertion failure (#15032)

### What this PR does / why we need it?

This PR fixes the following Triton Ascend assertion failure in the Mamba
state pre-copy kernel:

```text
PtrOffsetInfo::AxisInfo::operator[]: Assertion `idx < size()` failed
```
This issue was also observed in #7103 and #10888 .
This PR also:
- Adds an end-to-end test covering the supported layouts and execution
modes.
- Adds documentation for the kernel behavior, parameters, constraints,
and usage.
This is a follow-up to #14649.

After these changes, V2 Runner inference works correctly. 

- vLLM version: v0.27.1
- vLLM main:
https://github.com/vllm-project/vllm/commit/ba07e4a48fc951300d97eb506217dd530583dea3

---------

Signed-off-by: Lethobenthos20 <3205604419@qq.com>