Host decompression

[vuule]

Description

Add decompression APIs to make the nvCOMP use transparent.
Remove direct dependency on nvCOMP in the ORC and Parquet readers.
Add multi-threaded host-side decompression; currently off by default, can only be enabled via LIBCUDF_HOST_DECOMPRESSION environment variable.

Currently the host compression adds D2H + H2D transfers. Avoiding the extra transfers requires large changes to the readers to avoid device reads of the inputs.

Other changes:

• Replaced the host compression thread pool with the use of host_worker_pool, which is used for host decompression as well.
• Fixed logic in use_host_compression.
• Expanded host compression tests.
• Expanded error messages in compression.
• Allocate scratch space in gpu_debrotli to make the API consistent with other decompression types.
• Added GZIP to compress_max_allowed_chunk_size and compress_max_output_chunk_size to be able to use these functions for host GZIP compression.

Checklist

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

[copy-pr-bot]

Auto-sync is disabled for draft pull requests in this repository. Workflows must be run manually.

Contributors can view more details about this message here.

[vuule]

/ok to test

[vuule]

/ok to test