Implement fused matrix multiplication and ReLU in Triton. Input matrices $A$ and $B$ have shapes $(M,K)$ and $(K,N)$, and the output $Y$ has shape $(M,N)$:

Here, $M$ and $N$ are output dimensions, $K$ is the shared inner dimension, and $m$, $n$, and $k$ index those dimensions. Float32 denotes the arithmetic precision, and the maximum applies ReLU. Complete only the empty kernel body; the supplied solve launcher is implemented. The pointer arguments identify the three buffers, the size arguments give M, N, and K, the stride arguments give each buffer's element strides, and BLOCK_M, BLOCK_N, and BLOCK_K are compile-time tile sizes. Return no value; write Y into the supplied output tensor using its preallocated dtype.