[kernel] add AscendC op: lightning_indexer and sparse_flash_attention (#4625)

### What this PR does / why we need it?
Provide high-performance AscendC operators lightning_indexer and
sparse_flash_attention to boost the execution performance of the
DeepSeek v3.2 model. Meanwhile, adapt the two AscendC operators to
vllm-ascend framework.

### Does this PR introduce _any_ user-facing change?
No (only underlying operator optimizations, with no user-facing changes)

### How was this patch tested?

- vLLM version: v0.11.2
- vLLM main: https://github.com/vllm-project/vllm/commit/v0.11.2

Signed-off-by: MingYang119 <songmingyang@huawei.com>

vllm_ascend/attention/sfa_v1.py

@@ -459,7 +459,7 @@ class AscendSFAImpl(MLAAttentionImpl):
                                            kv_cache=kv_cache,
                                            attn_metadata=attn_metadata,
                                            need_gather_q_kv=need_gather_q_kv)
-        attn_output = torch.ops.custom.npu_sparse_flash_attention(
+        attn_output = torch.ops._C_ascend.npu_sparse_flash_attention(
             query=ql_nope,
             key=k_nope,
             value=k_nope,
@@ -554,7 +554,7 @@ class AscendSFAImpl(MLAAttentionImpl):
         seq_lens = attn_metadata.seq_lens
         cum_query_lens = attn_metadata.cum_query_lens
 
-        topk_indices = torch.ops.custom.npu_lightning_indexer(
+        topk_indices = torch.ops._C_ascend.npu_lightning_indexer(
             query=q,
             key=kv_cache[2],
             weights=weights,