fix: harden attention kernel boundaries

- fix scalar prefill head_dim=32 out-of-bounds via G=4 dispatch
- fix MMA decode 4D mask head indexing and invalid-row mask access
- add q_head/kv_head divisibility and head-dim contiguity checks
- validate split-KV scratch and decode out_buf layout in bindings
- set max dynamic shared memory for scalar decode D=256
- cover scalar prefill D=32 in pure C test

csrc/kernels/attn_decode_split_kv_mma.cuh

@@ -107,10 +107,11 @@ __global__ void attn_decode_split_kv_mma_kernel(AttentionParams<bf16> p) {
         int maxc = IsCausal ? KV::decode_attend_len(p, batch) : seq_len;
         mma_softmax_tile<Traits, HasMask>(kv0, maxc, maxc,
                                            0, 0,
-                                           p.mask_b_stride, 0, 0,
-                                           batch, 0,
-                                           p.mask,
-                                           Sacc, Oacc, m0, m1, l0, l1, lane);
+                                            p.mask_b_stride, p.mask_h_stride, p.mask_q_stride,
+                                            batch, q_head0 + gid, q_head0 + gid + 8,
+                                            p.mask,
+                                            va, vb,
+                                            Sacc, Oacc, m0, m1, l0, l1, lane);
 
         mma_pv_accumulate<Traits>(Sacc, bV, lane, Oacc);
     };