Remove CK BF16 gemm (#4851)

Summary:
Pull Request resolved: #4851

X-link: facebookresearch/FBGEMM#1877

This kernel [isn't used anywhere](https://www.internalfb.com/code/search?q=repo%3Afbcode%20torch.ops.fbgemm.bf16_gemm), and since we rely on rocBLAS for BF16, we probably don't need to keep this one now. It would be one less kernel to migrate over, and we plan to move the current `quantize` namespace to `gemm`.

Reviewed By: jwfromm

Differential Revision: D82114437

fbshipit-source-id: 7ec70a529613b2d69e278a6441ba66d9f57087de

Author: cthi

fbgemm_gpu/experimental/gen_ai/bench/ck_bf16_bench.py

@@ -37,9 +37,6 @@
     torch.ops.load_library(
         "//deeplearning/fbgemm/fbgemm_gpu/experimental/gen_ai:comm_ops"
     )
-    torch.ops.load_library(
-        "//deeplearning/fbgemm/fbgemm_gpu/experimental/gen_ai:gemm_ops"
-    )
     torch.ops.load_library(
         "//deeplearning/fbgemm/fbgemm_gpu/experimental/gen_ai:quantize_ops"
     )