[Apple Silicon] Add custom Metal RoPE kernel with fused KV cache store (#22868)

Signed-off-by: Xiaodong Ye <yeahdongcn@gmail.com>
Co-authored-by: Xiaodong Ye <yeahdongcn@gmail.com>

docs_new/docs/hardware-platforms/apple_metal.mdx

@@ -59,6 +59,7 @@ SGLANG_USE_MLX=1 python -m sglang.launch_server \
 1. `SGLANG_USE_MLX=1` - Enables the use of MLX as the SGLang runtime backend (if disabled, SGLang will fall back to `torch.mps`, which has less support)
 2. `--disable-cuda-graph` - Disables usage of CUDA graph, which is not relevant for Apple Metal.
 3. `--disable-overlap-schedule` - Disables overlap scheduling (enabled/not present by default) achieved using MLX's `async_eval()`
+4. `SGLANG_MLX_USE_CUSTOM_ROPE=1` - Enables the optional custom Metal RoPE kernel. It is disabled by default, so the MLX backend uses the standard RoPE path unless you opt in for A/B testing.
 
 ## Quantization
 
@@ -84,7 +85,6 @@ The MLX backend supports two quantization paths on Apple Silicon:
    ```
    The MLX backend silently ignores `--quantization mlx_q4` when the model is already quantized in its HF config (path 1), so the same flag is safe to pass either way.
 
-
 ## Benchmarking with Requests
 
 `sglang.benchmark_one_batch` calls the synchronous prefill/decode methods directly without going through the scheduler and the overlap code path.