cuda: decode expert groups take the grouped kernels even under TC_W4A16 (#431)

The TC_W4A16 branch of coli_cuda_expert_group handled every expert in a
per-expert loop: rows >= threshold got the Tensor Core path, everything
below fell back to 4 naive launches per expert. At decode every expert
has 1 row, so the whole group rode the fallback — ~981 quant_matmul
micro-launches per token (#431's measured flood) — while the grouped
3-launch path (grouped_hidden_w4_dual + silu + grouped_down_w4) sat one
else-if below, unreachable whenever the PREFILL tuning flag was set.

Gate the branch on 'at least one expert reaches the TC row threshold':
all-small groups (decode) now fall through to the grouped kernels.
Measured on 6x RTX 5090 (full residency, 39 forwards under nsys):
expert-side quant_matmul instances drop 981 -> 337 per forward, total
launches ~1,490 -> ~850 per token (-43%). Wall-clock is parity at the
A/B operating point — the win is structural (PR-C graph node count,
launch-tax share at champion speed).

Behavioural fix folded in: before this change, toggling TC_W4A16 — a
prefill-only optimization — changed DECODE output text (kernel-family
divergence, #100 class). After it, decode always uses the grouped
family: TC_W4A16=1 and =0 now produce byte-identical decode text
(verified, 96-token greedy A/B), and the flag affects only the prefill
it was built for.

c/backend_cuda.cu

@@ -689,7 +689,15 @@ extern "C" int coli_cuda_expert_group(ColiCudaTensor *const *gates,
         quantize_s4_rows<<<total,256,0,ctx->stream>>>(ctx->qx,ctx->qscale,ctx->gate,total,I);
         grouped_s4_wmma<<<dim3((unsigned)((D+63)/64),(unsigned)count),256,0,ctx->stream>>>(ctx->y,ctx->qx,ctx->qscale,dev,I,D,2);
     }else if(all_s4&&ctx->compute_major>=7&&getenv("COLI_CUDA_TC_W4A16")&&
-             atoi(getenv("COLI_CUDA_TC_W4A16"))){
+             atoi(getenv("COLI_CUDA_TC_W4A16"))&&
+             [&]{ int tc16_min=getenv("COLI_CUDA_TC_W4A16_MIN")?atoi(getenv("COLI_CUDA_TC_W4A16_MIN")):16;
+                  for(int c=0;c<count;c++) if(rows[c]>=tc16_min) return 1;
+                  return 0; }()){
+        /* At least one expert has enough rows for a Tensor Core tile. Groups
+         * where EVERY expert is below the threshold (decode: r=1) fall through
+         * to the grouped-W4 path below — 3 launches for the whole group instead
+         * of 4 per expert (#431: the launch flood measured at ~981 micro-kernels
+         * per token came from decode riding this branch's per-expert fallback). */
         /* W4A16 Tensor Core per gruppo: attivazioni fp16 per tile (lossless al
          * contrario del path W4A4), un lancio per expert dentro lo stream —
          * l'overhead di lancio e' trascurabile rispetto ai GEMM. */