Remove parquet decode type switch from loop, replace with if-constexpr in loop

[pmattione-nvidia]

This lifts the switch/if's on the column type out of the decode loop in the parquet fixed-width decode kernel. Prior to looping we instead identify the type and assign it an enum, and in the loop use if-constexpr's instead. This improves the performance of the decode.

Benchmark Results

• Non-chunked Int: 6-9% Faster
• Non-chunked float: 2-14% Faster
• Chunked Int: 10% Slower
• Chunked float: No change

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.