Add clang-tidy correctness CI

[scal444]

Updates to use clang-tidy 22 which has limited CUDA support.

We're starting with correctness-only items, then will expand with coverage and fixes .

Only true bug fixed in this diff is an unlikely overrun adding 1 to maxBatchSize BEFORE converting to size_t instead of after. The rest are defensive, or needed to get Clang CUDA compiling.

[greptile-apps]

Confidence Score: 5/5

The PR appears safe to merge; the previous checkout pinning concern is fixed and no actionable new failures were identified.

Summary

• Adds CUDA compile-database conversion and host-only CUDA analysis.
• Pins the CI checkout action to a reviewed commit SHA.
• Fixes size arithmetic so conversion occurs before incrementing maxBatchSize.
• Applies defensive initialization, explicit raw shared storage, and analyzer-oriented code refinements.

_{Reviews (2) · Last reviewed commit: "Add clang-tidy correctness CI"}