[Perf] Release the GIL in compute-intensive function

[haochengxia]

#59

Fixed by release GIL.

[INFO]  05-17-2026 00:42:47 cli_reader_utils.c:213  (tid=129999368357568): detecting trace type: ORACLE_GENERAL_TRACE
Thread A starting simulation...
[INFO]  05-17-2026 00:42:47 cli_reader_utils.c:213  (tid=129999034611392): detecting trace type: ORACLE_GENERAL_TRACE
Thread B starting simulation...
Thread B completed in 0.52s
Thread A completed in 0.54s

Total elapsed time: 0.57s

[gemini-code-assist]

Code Review

This pull request implements Global Interpreter Lock (GIL) management in the C++ cache export layer to enable multi-threaded trace processing from Python. Key changes include adding py::gil_scoped_acquire to cache operations that interact with Python objects and utilizing py::call_guard<py::gil_scoped_release> in the Python bindings for long-running simulation tasks. A new test suite, tests/test_gil.py, was added to verify concurrent execution. Review feedback highlights the need to acquire the GIL during cache destruction in pypluginCache_free to avoid crashes when cleaning up Python-owned parameters and suggests optimizing pypluginCache_get by delaying GIL acquisition until after the core C++ lookup is complete to reduce thread contention.