[naveenvenk17]

I found this while checking the default-off results for #11899.

TileGrid::initTiles() resizes the tile storage but keeps appending to its pointer list. Repeating initialization leaves duplicate pointers; changing the grid size can also leave references to old storage. The duplicates distort RUDY's weighted congestion metric because its percentile sample sizes are rounded up.

Clear the list and reserve space before rebuilding it.

Tested on the GCP Linux VM, with test-result caching disabled:

• All three new cases fail on the unchanged production control and pass with the fix. They cover repeated initialization, growth/shrinkage, pointer uniqueness, and actual nonuniform weighted-congestion calculations.
• 100 shuffled repetitions passed: 300 case executions.
• All seven repository formatting/lint/duplicate-ID checks passed.
• The full CPU GPL suite executed 110 targets: 109 passed. The remaining mt_invariance01 failure is an existing hardware INFO-line/golden-log mismatch. Its raw log, diff and DEF are byte-identical to the pure-upstream control, and the DEF matches the unchanged golden. Four GPU targets were skipped.

No golden files or metric limits were changed. This separates the grid bug from the opt-in inflation work in #11899.

[gemini-code-assist]

Code Review

This pull request ensures that the tiles_ vector is cleared and reserved during TileGrid::initTiles initialization to prevent stale pointers. It also introduces a comprehensive set of unit tests in route_base_test.cpp to verify grid reinitialization, resizing, and weighted congestion calculations. I have no feedback to provide.

[naveenvenk17]

I also ran all three new cases under ASan on the VM at bd6324e9. They passed with test-result caching disabled. Leak detection was off, as in the repo's ASan configuration.