plan: add confidence ratings + lift tasks 11-12 above 90%

Comply with the standing convention (memory dde30588 / 413d47550e):
every implementation-plan task carries a percentage confidence rating;
sub-95% tasks carry inline step-note annotations with mitigation;
sub-90% tasks must be lifted before execution or surfaced explicitly.

Lift passes applied:

- Task 11 (Launch phase): 88% -> 94%. Split consumeStockFor out of
  applyLaunches so Final Retaliation can fire from a corpse without
  push-stock-back. applyLaunches no longer gates on attacker.alive.

- Task 12 (Final Retaliation): 80% -> 91%. Removed the push-stock-back
  kludge. FR consumes stock once during launch synthesis, then hands the
  Launch[] directly to applyLaunches.

Also adds Confidence summary section with full task table, lift banner,
and four pre-merge executor checks (Math.random/Date.now/ui-import grep
gates plus determinism-test timing check).

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>

Author: emp3thy