Purpose

Linked issue: N/A

Enable ThreadSanitizer coverage, fix the data races exposed by the full test suite, and streamline the compiler and sanitizer CI workflows.

This change:

• separates the ASAN, UBSAN, and TSAN build switches while preserving the existing ASAN+UBSAN job;
• instruments bundled dependencies with TSAN, including building Boost with the configured C++ compiler and external-project compile/link flags;
• disables Tantivy in the TSAN job because its Rust static library is not TSAN-instrumented, and suppresses reports originating from prebuilt Lance, Lumina, and Jindo shared libraries;
• makes compaction metric snapshots and test cache counters safe under concurrent access;
• consolidates GCC 14 and Clang Release/Debug builds into a single matrix workflow while keeping GCC 8 independent;
• uses concise sanitizer job names and named build-script arguments;
• compares clang-tidy changes against the pull request target branch instead of always using origin/main;
• normalizes external-project flag propagation and preserves the warning-flag ordering required by Snappy; and
• prints captured third-party build logs when an ExternalProject step fails.

Tests

• Clang 17 TSAN full build with Jindo, Lance, Lumina, and Lucene enabled; Tantivy disabled as configured by the TSAN CI job.
• ctest --test-dir build-pr-final --output-on-failure -j 64 (27/27 passed).
• Clang 17 Release build of the bundled snappy_ep target.
• pre-commit run --files <all changed files>.
• bash -n ci/scripts/build_paimon.sh.
• git diff upstream/main...HEAD --check.

API and Format

No API, storage format, or protocol changes.

Documentation

No user-facing documentation changes.

Generative AI tooling

Generated-by: Codex (GPT-5)