[NPU] support mtp(beta) pd disaggregation and dp attention

[iforgetmyname]

…raph(npu) & support dsv3_2 mtp

Motivation

Modifications

Accuracy Tests

Benchmarking and Profiling

Checklist

• Format your code according to the Format code with pre-commit.
• Add unit tests according to the Run and add unit tests.
• Update documentation according to Write documentations.
• Provide accuracy and speed benchmark results according to Test the accuracy and Benchmark the speed.

[sglang-bot]

1. upload torch profile to show that the overlap actually happens and there is no CPU overhead.
   • Rule: If you change any logic in (overlap) scheduler, attach a torch profile to show the overlap actually happens.
2. compare the speed and acceptance length of overlap vs. non-overlap
3. Get someone to verify on GPU
4. add a test case

[sglang-bot]

cc @ShangmingCai @hnyls2002

[sglang-bot]

ways to remove if/else

• Pu all if else into a single place (ideally in the initialization)
• Try to reuse batch.input_ids.device or batch.sampling_info

[sglang-bot]

Device2ExtendCudaGraphRunner = {
    "npu": EAGLEDraftExtendNpuGraphRunner,
    "cuda": EAGLEDraftExtendCudaGraphRunner,
}