docs: Add FP16 GEMM documentation to sgemm_sm80.cu - Fixes #1686

[blueberrycongee]

Summary

Add descriptive documentation to sgemm_sm80.cu explaining that it is actually an FP16xFP16 GEMM (HGEMM) tutorial using CuTe.

Problem

Issue #1686 reports that users cannot find an fp16 GEMM tutorial in the CuTe examples.

Solution

The existing sgemm_sm80.cu already implements FP16 GEMM using cute::half_t, but this was not documented. This PR adds a documentation block clarifying:

• This example uses FP16 data types despite the "sgemm" filename
• Key features: Tensor Cores, cp.async, pipelining, swizzled shared memory
• Usage examples

Fixes #1686

[hwu36]

if the accumulation type is fp32, it is sgemm. if the accumulation type is fp16, it is hgemm.