refactor: rename gqa_* to attn_*, split-KV for all decode paths

- Rename all csrc/kernels/gqa_*.cuh/cu to attn_*, with _split_q / _split_kv
  strategy suffix and optional _mma compute suffix
- Remove non-split MMA decode kernel, keep only split-KV path
- Convert scalar decode fallback to split-KV (o_part/ml_part + combine)
- Move combine kernel to attn_decode_split_kv.cuh (shared by both paths)
- Rename GQAParams to AttentionParams
- Update all C++ #include, PYBIND11, and Python extension references

astrai/extension/__init__.py

@@ -1,19 +1,19 @@
 """CUDA attention kernel wrappers with torch fallback.
 
 Public API:
-    - ``gqa_decode_attn`` — single-query decode attention
-    - ``gqa_prefill_attn`` — multi-query prefill attention
+    - ``attn_decode`` — single-query decode attention
+    - ``attn_prefill`` — multi-query prefill attention
 
-Each wrapper dispatches to its compiled CUDA kernel (``astrai.extension.gqa_*``)
+Each wrapper dispatches to its compiled CUDA kernel (``astrai.extension.attn_*``)
 when available, otherwise falls back to ``torch.nn.functional.scaled_dot_product_attention``.
 """
 
 from astrai.extension.loader import KERNEL_NAMES, is_available
-from astrai.extension.ops import gqa_decode_attn, gqa_prefill_attn
+from astrai.extension.ops import attn_decode, attn_prefill
 
 __all__ = [
-    "gqa_decode_attn",
-    "gqa_prefill_attn",
+    "attn_decode",
+    "attn_prefill",
     "is_available",
     "KERNEL_NAMES",
 ]

astrai/extension/loader.py

@@ -11,7 +11,7 @@ import logging
 
 logger = logging.getLogger(__name__)
 
-KERNEL_NAMES = ["gqa_decode_attn", "gqa_prefill_attn"]
+KERNEL_NAMES = ["attn_decode", "attn_prefill"]
 
 _available: dict[str, bool] = {}
 _modules: dict[str, object] = {}

astrai/extension/ops.py

@@ -42,7 +42,7 @@ def _torch_fallback(
     )
 
 
-def gqa_decode_attn(
+def attn_decode(
     q: torch.Tensor,
     k: torch.Tensor,
     v: torch.Tensor,
@@ -51,8 +51,8 @@ def gqa_decode_attn(
     causal_offset: int = 0,
     scale: float | None = None,
 ) -> torch.Tensor:
-    if _available["gqa_decode_attn"]:
-        return _modules["gqa_decode_attn"].gqa_decode_attn(
+    if _available["attn_decode"]:
+        return _modules["attn_decode"].attn_decode(
             q,
             k,
             v,
@@ -64,7 +64,7 @@ def gqa_decode_attn(
     return _torch_fallback(q, k, v, mask, is_causal, scale)
 
 
-def gqa_prefill_attn(
+def attn_prefill(
     q: torch.Tensor,
     k: torch.Tensor,
     v: torch.Tensor,
@@ -73,8 +73,8 @@ def gqa_prefill_attn(
     causal_offset: int = 0,
     scale: float | None = None,
 ) -> torch.Tensor:
-    if _available["gqa_prefill_attn"]:
-        return _modules["gqa_prefill_attn"].gqa_prefill_attn(
+    if _available["attn_prefill"]:
+        return _modules["attn_prefill"].attn_prefill(
             q,
             k,
             v,