Race condition in multi-threaded syrk calls

[lujiaweics]

Summary

Concurrent multi-threaded calls to syrk cause a race condition due to missing mutex locks and condition variables. This may lead to incorrect results or crashes.

The same synchronization logic (locks + condition variables) is already implemented for:

• GEMM: level3_thread.c
• GEMM3M: level3_gemm3m_thread.c

A corresponding fix has been submitted in MR #5619.