fix: GEMV NaN bug — skip custom kernel for small N (<256)

The custom launch_gemv_bf16 kernel produces NaN when output dimension N
is small (e.g. N=32 for the MoE router). Fall back to cuBLAS GemmEx for
N < 256. Also removes the padding workaround in gpt_oss MoE forward.

Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>

crates/xserv-kernels/src/gemm.rs

@@ -173,7 +173,7 @@ pub fn matmul(a: &Tensor, b: &Tensor, backend: GemmBackend) -> Tensor {
             }
         }
         GemmBackend::CuBlas => {
-            if m == 1 && dtype == DType::BF16 {
+            if m == 1 && dtype == DType::BF16 && n >= 256 {
                 let mut fp32_buf = xserv_cuda::allocator::cached_alloc(n * 4).unwrap();
                 unsafe {
                     launch_gemv_bf16(