Make is_decompression_disabled and is_compression_disabled thread-safe

[vuule]

Description

Adds a mutex to guard the use of the memoization map for the status of nvCOMP features.

Checklist

• I am familiar with the Contributing Guidelines.
• New or existing tests cover these changes.
• The documentation is up to date with these changes.

[ttnghia]

I wonder if we can just call is_(de)compression_disabled_impl directly without the caching mechanism, as these functions are very cheap.

[vuule]

I wonder if we can just call is_(de)compression_disabled_impl directly without the caching mechanism, as these functions are very cheap.

Yes, we could. I only added this mechanism to be able to log the result of each query once. Otherwise we would either need a similar map for logging, or we would have to allow logs to be very repetitive.

[abellina]

@vuule I ran with this and it did not reproduce. Thanks for the quick fix.

[abellina]

We are hitting all kinds of segfaults in CI that I think are all related to this issue: NVIDIA/spark-rapids#8194, so the spark-rapids CI is currently blocked or limping along. Could we merge @vuule's solution and then perhaps follow up with more modifications if the memoization approach should be modified?

[ttnghia]

/merge