[FEA] Add multi-threaded GZIP decompression for multi-source JSON reading

[GregoryKimball]

Is your feature request related to a problem? Please describe.
The JSON reader in libcudf supports multi-source reading of GZIP-compressed JSONL files, using host-side decompression algorithms.

However, the performance is limited to about 100 MB/s due to a single-host thread completing the decompression in sequence (see discussion in 17219).

Describe the solution you'd like
We should add a multi-threaded implementation to process GZIP decompression, with one host thread per source. Each source is a single compression block.