[Kernel] Add AscendC fused op transpose_kv_cache_by_block to speed up GQA transfer (#6366)

### What this PR does / why we need it?
As #2947 describe, we need to transpose kv cache layout after GQA kv
transfer when prefill and decode tensor parallel size are heterogeneous,
in the previous implementation, we use `npu_paged_cache_load ` +
`tranpose` + `_npu_reshape_and_cache` to do this work.

But obviously, it is not an efficient plan, the ops above need to be
called for each layer, which introduces 3 * layer_num kernel launch, and
6 * layer_num data movement between L1 Cache and HBM for one request on
decode node. Usually, decode node uses graph mode, so these op kernels
will be called between decode forward launched by an async thread in
mooncacke connector, this kernels maybe last for several decode forward
and TTFT will increase by 3~4 decode forward time.

In this PR, we implement an AscendC fused op
`transpose_kv_cache_by_block` to do this with only once kernel launch
and move data between L1 Cache and HBM only once.

After using this fused op, the time cost in transpose kv cacke layout
can be decreased to 0.24ms from 7ms in UT on 910C, and in PD
disaggregation scenario, TTFT can decrease about 90 ~ 110 ms in
qwen3-235B.

| request_num | original | fused_op|
|:----------------------:|:---------------:|:-------------------:|
|           1            |      643 ms      |        578 ms        |
|          128           |     1480 ms      |       1368 ms        |

### Does this PR introduce _any_ user-facing change?
Use fused op by default, incase the op has bug in any scenario, provide
fallback choice using env to disable it.

**DISABLE fused op by add following env**
`export VLLM_ASCEND_FUSION_OP_TRANSPOSE_KV_CACHE_BY_BLOCK=0`

### How was this patch tested?

- vLLM version: v0.14.1
- vLLM main:
dc917cceb8

---------

Signed-off-by: lidenghui <lidenghui1110@gmail.com>

csrc/torch_binding.cpp

@@ -1343,6 +1343,22 @@ std::tuple<at::Tensor,at::Tensor, at::Tensor> npu_add_rms_norm_bias(
     return std::tuple<at::Tensor, at::Tensor, at::Tensor>(y, rstd, x);
 }
 
+void transpose_kv_cache_by_block(
+    const at::TensorList &kCache,
+    const at::TensorList &vCache,
+    const at::Tensor &blockIDs,
+    int64_t blockSize,
+    int64_t headNum,
+    int64_t headDim,
+    int64_t splitNum,
+    int64_t layerNum)
+{
+
+    EXEC_NPU_CMD(aclnnTransposeKvCacheByBlock, kCache, vCache, blockIDs,
+                 blockSize, headNum, headDim, splitNum, layerNum);
+
+}
+
 } // namespace vllm_ascend
 
 TORCH_LIBRARY_EXPAND(CONCAT(_C, _ascend), ops)
@@ -1521,4 +1537,8 @@ TORCH_LIBRARY_EXPAND(CONCAT(_C, _ascend), ops)
 
     ops.def("npu_apply_top_k_top_p(Tensor logits, Tensor? p=None, Tensor? k=None) -> Tensor");
     ops.impl("npu_apply_top_k_top_p", torch::kPrivateUse1, &vllm_ascend::npu_apply_top_k_top_p);
+    ops.def(
+        "transpose_kv_cache_by_block(Tensor[] kCache, Tensor[] vCache, Tensor blockIDs, int blockSize, int headNum, int headDim, int splitNum, int layerNum) -> ()"
+    );
+    ops.impl("transpose_kv_cache_by_block", torch::kPrivateUse1, &vllm_ascend::transpose_kv_cache_by_block);
 }