feat: initial vllm-npu-plugin for Ascend NPU adaptation

- NPUPlatform: device management, HCCL process group, config adaptation
- AscendAttentionBackend: npu_fusion_attention (prefill) + npu_incre_flash_attention (decode)
- NPUCommunicator: HCCL-based distributed communication
- NPUWorker: NPU device init, memory profiling
- Custom ops: SiluAndMul, RMS norm, rotary embedding
- Plugin registered via vllm.platform_plugins entry point

Based on vllm-ascend official pattern, targeting Ascend 910B

vllm_npu/ops/activation.py

@@ -0,0 +1,17 @@
+"""
+NPU-optimized activation functions for Ascend.
+
+Provides ``AscendSiluAndMul`` that uses ``torch_npu.npu_swiglu`` for
+fused SiLU+Mul on NPU devices.
+"""
+
+import torch
+from vllm.model_executor.layers.activation import SiluAndMul
+
+
+class AscendSiluAndMul(SiluAndMul):
+    """SiluAndMul using torch_npu.npu_swiglu on Ascend NPU."""
+
+    def forward_oot(self, x: torch.Tensor) -> torch.Tensor:
+        import torch_npu  # noqa: F401
+        return torch_npu.npu_swiglu(x)