Re-measure chunk-only overhead and trim benchmarks to the evaluated set

[byeongjee]

Summary

• Re-measure results/chunking_overhead/chunked.csv at 10uF after recent pass changes (execution times within ~±4% of the previous run)
• Remove the 12 benchmarks outside the 13-benchmark evaluated set (*_na, basicmath, ecc, fft, pathological_milp_vs_schematic, test) and the now-dead test exclusion in plot_results.R
• Remove leftover debugging .ll / .ll.tmp files from benchmarks/intermittent/

Test plan

• uv run ckpt bench chunked --cap 10uF on device
• Pre-commit hooks (ruff, pyright) pass