eval(debugging): close post-improvement re-measure of full debugging cluster

Re-ran five-whys-plus and occams-razor on SWE-bench (n=150, isolated,
length-controlled) after the in-place audit edits, completing the
before/after table for all four native-debugging skills.

Results (placebo -> with-skill):
- five-whys-plus (trimmed): +4.0pp p=0.041 SIG -> +3.3pp p=0.131 ns
- occams-razor (trigger-scoped): +2.0pp p=0.45 ns -> +2.0pp p=0.505 ns

Findings:
- Two independent trimming regressions now (systems +5.3->+3.3,
  five-whys +4.0->+3.3): trimming a proven skill loses ~1-2pp. The
  description/boundary edits are safe; the content trimming is suspect.
- occams stayed flat: a cosmetic edit is not a rework. Only the
  agent-native rewrite (scientific-method, 0->+9.3pp p=0.002) moved it.

Records exp-eval-swe-{five-whys-plus,occams-razor}-improved to
experiments.db; updates ELEVATE-OR-KILL.md (full 4-skill table + 3
lessons) and progress.md. Closes task #32.

Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>