[Feat]: support multistream for CAMP2p

[yujuancao07]

Purpose

Add configurable multi-stream execution to the Ascend CAMP2P connector
so that A2F/F2A communication can overlap with model computation.

Attention-side and FFN-side multi-stream execution can be enabled independently
through connector_extra_config.

Issue

• Related issue(s): [Feature]: Enable multi-stream parallelism for CAMP2P A2E/E2A on FFN side #68
• Closing keyword, only if fully resolved: Closes [Feature]: Enable multi-stream parallelism for CAMP2P A2E/E2A on FFN side #68

Scope

This feature applies only to CAMP2pAFDConnector on NPU.

• In scope:

  • Attention-side CAMP2P communication stream execution.
  • FFN-side CAMP2P communication stream execution.
  • Independent Attention and FFN configuration switches.
  • Independent Attention and FFN AIV core-count configuration.
  • Eager execution and the existing AFD NPU ubatch/DBO path.
  • Existing ACL graph paths where stream switching can be represented safely by
    the current graph runtime.

• Out of scope:

  • CAMAsyncAFDConnector.
  • GPU P2pNcclAFDConnector.

Implementation Notes

• The implementation applies only to CAMP2pAFDConnector.

• Both switches are located under CAMP2P connector_extra_config:

  {
    "connector_extra_config": {
      "is_attn_multistream": true,
      "is_ffn_multistream": true,
      "attn_core_num": 8,
      "ffn_core_num": 8
    }
  }
  

Test Plan

Focused multi-stream tests:
pytest -q tests/unit -k multistream -vv

Test Result

[yujuancao07]

multistream:

sigle stream:

[mergify]

⚠️ The sha of the head commit of this PR conflicts with #146. Mergify cannot evaluate rules on this PR. Once #146 is merged or closed, Mergify will resume processing this PR. ⚠️