refactor: unify paged and contiguous attention kernels via KVSource policy

- merge AttentionParams and PagedAttentionParams into one struct
- add attn_kv_source.cuh with ContigKV/PagedKV addressing policies
- template prefill/decode kernels (MMA + scalar) on the KV policy, deleting the four duplicated attn_paged_*.cuh variants
- template dispatcher launchers on KV; single combine kernel
- verify: all correctness tests pass and SASS matches baseline (no perf regression)

csrc/kernels/attn_entry_utils.cuh

@@ -149,7 +149,7 @@ inline void attn_pack_paged_decode_params(
     c10::optional<torch::Tensor> mask,
     int64_t causal_offset,
     double scale,
-    PagedAttentionParams<T>& p
+    AttentionParams<T>& p
 ) {
     const at::cuda::OptionalCUDAGuard device_guard(device_of(q));
 
@@ -229,7 +229,7 @@ inline void attn_pack_paged_prefill_params(
     int64_t max_q_len,
     int64_t causal_offset,
     double scale,
-    PagedAttentionParams<T>& p
+    AttentionParams<T>& p
 ) {
     const at::cuda::OptionalCUDAGuard device_guard(device_of(q));