Remove cross-run duration weighting from test partitioning

[MatthewMckee4]

Summary

The runner used to read the previous run's per-test durations from the cache and weight the partitioning algorithm by those timings to balance worker load. In practice the added complexity wasn't paying off, so this drops the cross-run read entirely and partitions using uniform test counts instead. Module-aware bin-packing is preserved (tests in the same module stay on one worker when possible; large modules still split across workers) — only the duration-based weighting is gone. --durations N still works; per-worker durations are still written and aggregated within a run.

Test plan

ci

[codspeed-hq]

Merging this PR will degrade performance by 18.75%

⚠️ Unknown Walltime execution environment detected

Using the Walltime instrument on standard Hosted Runners will lead to inconsistent data.

For the most accurate results, we recommend using CodSpeed Macro Runners: bare-metal machines fine-tuned for performance measurement consistency.

❌ 1 regressed benchmark

⚠️ Please fix the performance issues or acknowledge them on CodSpeed.

Performance Changes

	Mode	Benchmark	BASE	HEAD	Efficiency
❌	WallTime	karva_benchmark	24.7 s	30.4 s	-18.75%

---

_{Comparing remove-duration-cache (73468bd) with main (faef1bf)}