Reduce local colcon test time

[bburda]

Summary

Local colcon test takes ~20 minutes because it runs everything serially: clang-tidy (~8-10 min), unit tests (~3 min), and integration tests (~7 min). During typical feature development, most of this is wasted - developers only need fast unit test feedback.

---

Proposed solution (optional)

Six incremental changes:

1. Developer convenience script (scripts/test.sh) - named presets (unit, integ, lint, tidy, all) with parallel CTest execution via --parallel-workers $(nproc)
2. Opt-in clang-tidy - gate behind ENABLE_CLANG_TIDY=OFF (default), CI sets it ON for Jazzy linter job
3. ccache integration - auto-detect ccache via cmake/ROS2MedkitCcache.cmake, include in all 5 C++ packages
4. Precompiled headers - PCH for rclcpp, nlohmann/json, httplib, common STL headers in gateway_lib (~30-40% build speedup)
5. CI ccache - persistent ccache via actions/cache for faster CI rebuilds on subsequent runs
6. Incremental clang-tidy - pre-push hook running clang-tidy only on changed .cpp files via merged compile_commands.json

Expected result: typical dev loop (edit + unit test) drops from minutes to under a minute with warm ccache.

---

Additional context (optional)

• Gateway package alone: 142 files / 38k LOC, clang-tidy TIMEOUT 1500s
• All changes are backward-compatible - no effect when ccache is absent, clang-tidy still mandatory in CI
• PCH can be disabled with -DCMAKE_DISABLE_PRECOMPILE_HEADERS=ON