Add online CUDA lookup and GPU/cuCollections benchmark scaffolding

[tpn]

Closes #78.

Summary

• add online-JIT CUDA export APIs for emitted CUDA source, table payloads, and table metadata
• add standalone GPU benchmark/examples for NVRTC-based lookup plus cuCollections static_map and static_multiset baselines
• add TPC-H query-probe extraction utilities for real build/probe stream benchmarking
• harden the new benchmarks and emitter paths based on repeated review feedback around verification, portability, runtime loading, and edge cases

Validation

• cmake --build build-online-llvm-jit --config Release --target PerfectHashOnlineCore -j 8
• python -m py_compile examples/tpch-query-probes/extract_tpch_query_probes.py examples/tpch-query-probes/partition_hot_subset.py
• cmake --build /tmp/ph-cuco-map-build -j 8
• cmake --build /tmp/ph-cuco-multiset-build -j 8
• cmake -S examples/cpp-console-online-cuda-nvrtc -B /tmp/ph-online-cuda-build -DPERFECTHASH_GIT_REPOSITORY=file:///home/trentn/src/perfecthash -DPERFECTHASH_GIT_TAG=78-online-cuda-gpu-bench -DPERFECTHASH_BUILD_PROFILE=online-llvm-jit >/dev/null && cmake --build /tmp/ph-online-cuda-build -j 8

Review

• repeated roborev branch reviews were run during the rebase/update cycle
• the final daemon-backed Codex review path was blocked by provider auth issues (401 Unauthorized on gpt-5.4), so the last clean review pass I used was via claude-code, followed by the remaining low-severity cleanup nits incorporated directly

[chatgpt-codex-connector]

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit: adc9fe4f5a

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".