chore(metrics): measure cheap-blocker abort, countable pre-phase and tiered findings (#2164)

[bdfinst]

Part of #2200. Blocked on slice 0.

Measure

• Abort (feat(code-review): abort remaining panel dispatches on a cheap-lens blocker #2168): abort frequency per checkpoint, the opus-tier tokens deferred, and how often the deferred lenses found something new after the fix loop. If deferred lenses almost always come back clean, the abort is pure savings. If they often find something, the abort is only reordering the work.
• Countable pre-phase (feat(agents): countable pre-phase in test-review, as a pilot #2169): mean test-review dispatch tokens before vs. after feat(review): abort-on-cheap-blocker, countable test-review, tiered findings #2197's merge (cost_meter / /cost-report), plus /agent-eval true-positive parity on the test-review fixtures. A drop in true positives fails this mechanism regardless of any token savings.
• Tiered findings (feat(code-review): tiered findings output with on-demand expansion #2170): extend the one-round 59.0% report-size measurement to N ≥ 5 rounds with render_tiered_findings.py, and count how often --expand is used. Frequent expansion would mean Tier-1 lines are too thin.

Acceptance criteria

• The before/after token delta is reported with the confounds pre-registered in slice 0, as directional rather than causal.
• Each of the three mechanisms is scored against its slice-0 threshold, with the instrument named, and posted to chore(metrics): evaluate realized benefits of epics #2164 and #2172 #2200.

---

Generated by Claude Code