refactor: unify extension API to blhd layout and is_causal

- Rename ops.py to attention_ops.py
- Remove layout/scale params: fixed blhd, auto scale
- Replace causal_offset with is_causal bool
- Move SDPA fallback to backend, ops only calls CUDA kernels
- Update __init__.py exports

astrai/extension/__init__.py

@@ -4,27 +4,39 @@ Public API:
     - ``attn_decode`` — single-query decode attention
     - ``attn_prefill`` — multi-query prefill attention
     - ``attn_paged_decode`` — paged decode attention (direct page-table access)
+    - ``AttentionBackend`` — ABC for attention computation strategies
+    - ``TorchNativeBackend`` — default SDPA backend with KV cache I/O
 
-Interface (shared by all wrappers):
-    causal_offset: -1 = non-causal; >=0 = absolute position of first Q token
-    mask:       2D [batch, kv_len] or 3D [batch, q_len, kv_len] (bool, True = keep)
-    scale:      0.0 = auto (1/sqrt(head_dim)); >0 = explicit
-    layout:     "bhld" (default) or "blhd"
-
-Causal and mask can coexist — both are applied simultaneously.
+Layout convention: all q/k/v are ``[batch, seq_len, n_heads, head_dim]``
+(blhd). Scale is always ``1/sqrt(head_dim)``.
 
 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.attention_backend import (
+    ATTN_BACKEND,
+    AttentionBackend,
+    TorchNativeBackend,
+    attn_backend,
+    get_backend,
+)
+from astrai.extension.attention_ops import (
+    attn_decode,
+    attn_paged_decode,
+    attn_prefill,
+)
 from astrai.extension.loader import KERNEL_NAMES, is_available
-from astrai.extension.ops import attention, attn_decode, attn_paged_decode, attn_prefill
 
 __all__ = [
+    "ATTN_BACKEND",
+    "AttentionBackend",
+    "TorchNativeBackend",
+    "attn_backend",
+    "get_backend",
     "attn_decode",
     "attn_paged_decode",
     "attn_prefill",
-    "attention",
     "is_available",
     "KERNEL_NAMES",
 ]