[CuTe][SM70] Add comment clarifying signed cast requirement for blockIdx coords

[Flink-ddd]

Problem

blockIdx returns uint3, so m_coord, n_coord, and l_coord are unsigned. When these are passed directly to make_coord, the tile residue calculations:

auto m_max_coord = size<0>(shape_MNKL) - size<0>(gA) * get<0>(cta_coord);
auto n_max_coord = size<1>(shape_MNKL) - size<0>(gB) * get<1>(cta_coord);

can produce negative values for small problem shapes (e.g. M=8, N=8 with TileShape=128x128). With unsigned arithmetic, these wrap around to large positive values, causing incorrect predication and wrong results.

Fix

The int() cast in make_coord was already present in sm70_gemm.hpp and sm70_gemm_array.hpp, but without explanation. This PR adds a comment to clarify why the cast is necessary, so users writing custom kernels based on these files do not accidentally omit it.

Reported in #3190.