[Unity][BYOC] Cache cuBlasLt handle with thread entry

[vinx13]

Added CuBlasLtThreadEntry to cache cublasLt handle. Without it, cudaDeviceSynchonize is inserted and causes overhead.

cc @masahi @yelite

[tvm-bot]

Thanks for contributing to TVM! Please refer to the contributing guidelines https://tvm.apache.org/docs/contribute/ for useful information and tips. Please request code reviews from Reviewers by @-ing them in a comment.

• cc @billishyahao, @quic-sanirudh _{See #10317 for details}

_{Generated by tvm-bot}

[masahi]

I get this compile error using g++11:

/home/masa/projects/dev/tvm/src/runtime/contrib/cublas/cublas_json_runtime.cc:45:23: error: using-declaration for non-member at class scope
   45 |   using tvm::contrib::CuBlasLtThreadEntry;
      |                       ^~~~~~~~~~~~~~~~~~

[masahi]

I removed using and replaced this line with

auto entry_ptr = tvm::contrib::CuBlasLtThreadEntry::ThreadLocal();

to workaround the compile error.

[masahi]

@vinx13 Need to fix lint.