Refactor dictionary encoding in PQ writer to migrate to the new cuco::static_map

[mhaseeb123]

Description

Part of #12261. This PR refactors the dictionary encoding in Parquet writers to migrate from cuco::legacy::static_map to cuco::static_map to build the dictionaries.

Performance Results

The changes result in +0.08% average speed improvement and +16.22% average memory footprint increase (stems from the adjusted sizes by cuco::make_window_extent due to prime gap) across the benchmark cases extended from #16591

Currently, we do see a roughly 8% speed improvement in map insert and find kernels which is counteracted by the map init and map collect kernels as they have to process 16.22% more slots. With a cuco bump, the average speed improvement will increase from 0.08% to +3% and the memory footprint change will go back from 16.22% to +0%.

Hardware used for benchmarking

 `NVIDIA RTX 5880 Ada Generation`
* SM Version: 890 (PTX Version: 860)
* Number of SMs: 110
* SM Default Clock Rate: 18446744071874 MHz
* Global Memory: 23879 MiB Free / 48632 MiB Total
* Global Memory Bus Peak: 960 GB/sec (384-bit DDR @10001MHz)
* Max Shared Memory: 100 KiB/SM, 48 KiB/Block
* L2 Cache Size: 98304 KiB
* Maximum Active Blocks: 24/SM
* Maximum Active Threads: 1536/SM, 1024/Block
* Available Registers: 65536/SM, 65536/Block
* ECC Enabled: No

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]

This pull request requires additional validation before any workflows can run on NVIDIA's runners.

Pull request vetters can view their responsibilities here.

Contributors can view more details about this message here.

[mhaseeb123]

/ok to test

[PointKernel]

Redundant capacity issue is tracked via NVIDIA/cuCollections#581

[mhaseeb123]

/ok to test

[PointKernel]

One small question otherwise looks good to me.

Great work!

[mhaseeb123]

/ok to test

[mhaseeb123]

/merge