refactor: rework attention backend resolution

- explicit attn_backend() context wins over ASTR_BACKEND env
- polymorphic available()/supports_call() replace isinstance dispatch
- cache singleton backend instances to avoid hot-path allocation
- training (fwd=None) resolves cuda > flash > torch by capability
- flash dense supports mask-free calls only; masked training falls back to torch

docs/developer/internals.md

@@ -188,7 +188,7 @@ Attention computation is decoupled from the model via `AttentionBackend` ABC (`a
 - **`FlashAttnBackend`**: optional flash-attn dispatch with `flash_attn_with_kvcache` fast path for contiguous cache; falls back to KV gather + `flash_attn_func`.
 - **`TorchNativeBackend`** (always-available fallback): writes K/V to cache, gathers via `req_to_token` indirect indexing, calls `F.scaled_dot_product_attention`.
 - The `attention(...)` entry point uses cuda > flash > torch priority and chooses another compatible backend when an automatically selected backend cannot handle a call.
-- `ASTR_BACKEND=cuda|torch_native|flash` and `attn_backend(...)` are explicit selections; incompatible calls raise instead of silently changing backend.
+- Resolution precedence is: explicit `attn_backend(...)` context > `ASTR_BACKEND` env > default. An explicit `attn_backend(...)` selection is strict (incompatible calls raise); `ASTR_BACKEND` is a default-level override that falls back to a compatible backend when incapable. Training calls (`fwd=None`, no KV cache) resolve by capability: the CUDA cache kernels cannot run without a cache, so they fall back to flash (mask-free/causal calls only) and finally to torch SDPA.
 
 Rotary embedding is applied via `apply_rotary_emb` in `astrai/extension/backend/rotary.py`, which auto-dispatches to the fused CUDA kernel (`rotary_emb.cu`) during inference or torch complex multiply during training (for autograd compatibility). Both attention backends share the same rotary dispatch.