perf: move decode split partials to InferenceWorkspace

- Replace per-.cu-file static cached tensors with workspace-managed pre-allocated buffers

- InferenceWorkspace now owns decode_o_part / decode_ml_part (mirrors FlashInfer's workspace pattern)

- KVCache carries the buffers through the backend -> C++ kernel chain

- C++ kernels accept optional pre-allocated buffers; fallback to alloc_split_partials for backward compat

- Pre-allocates once at Executor init, zero allocation in the decode hot loop

- Prerequisite for CUDA-graph capture (all kernel addresses are stable)

astrai/extension/attention_backend.py

@@ -441,6 +441,8 @@ class CudaBackend(AttentionBackend):
             kv_indptr,
             kv_cache.max_len,
             is_causal=True,
+            o_part_buf=kv_cache.decode_o_part,
+            ml_part_buf=kv_cache.decode_ml_part,
         )
         return out.unsqueeze(1).flatten(2)