perf: add fused CUDA rotary embedding kernel

- Single-kernel rotary embedding (cos/sin lookup + rotation) replaces PyTorch complex-multiply path (3 kernel launches + f32 upcast per call)
- RotaryEmbedding now stores cos_table/sin_table and returns (cos, sin) f32 tuple instead of a complex tensor
- apply_rotary_emb in rotary_backend.py auto-dispatches: CUDA kernel if available, else torch complex-multiply fallback; backend-agnostic (both attention backends benefit)
- Kernel: 256-thread blocks, grid-stride loop, vectorized __nv_bfloat162 load/store, f32 compute, bf16 out
- Standalone kernel 6-9x faster than torch across decode/prefill shapes, max diff 0 (decode) to 3e-2 (large prefill, bf16)
- Benchmark (L20, bf16, CUDA backend): B=1 9.48->7.25ms (+31%), B=4 10.73->7.67ms (+40%), B=8 10.77->7.81ms (+38%), B=16 10.79->7.83ms (+38%)

astrai/extension/loader.py

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