feat(agent): add host readiness audit receipts

[GTC2080]

Summary

• Adds runtime-only readiness audit receipts for the existing final-answer readiness gate.
• Records blocked, allowed, recovered, and terminal-error outcomes without changing the gate policy.
• Extends run --metrics JSON with readiness counters for blocks, recoveries, errors, missing project checks, missing complete_step, and command mismatch/missing cases.

Test Plan

• D:\Go\go\bin\go.exe test ./internal/evidence ./internal/event ./internal/agent ./internal/cli
• D:\Go\go\bin\go.exe test ./internal/...
• D:\Go\go\bin\go.exe test ./...
• git diff --check

Review Evidence

Related to #2830

No UI changes; screenshots not applicable.

No performance claims; cache hit rate/token/runtime improvements are not claimed.

No prompt or tool schema changes.

e2e was not triggered from this account; the workflow is trusted-author gated. A maintainer can trigger /e2e diff x3 if desired.

[esengine]

I built this and ran the full readiness-audit suite locally — all green (block+recovery, terminal error, skip-when-not-applicable, metrics accumulation/write fields, event-sync forwarding), and the existing readiness gating tests still pass. The behavior is solid.

It needs a rebase before it can land, though: the final-readiness gate moved under you after this branched. #2920 (gate final answers on incomplete todos) and #2933 (allow plan-mode todos in final readiness) reworked finalReadinessFailure, and internal/cli/run_metrics.go also diverged — so rebasing onto main-v2 conflicts in both. Because the conflict is in the core gate logic (your struct refactor vs. the new missing-checks logic now on main), I'd rather you reconcile it than have me guess at the merge.

Please git rebase origin/main-v2, fold the current gate's logic into your finalReadinessCheck struct (so the audit records the new conditions too), and push. Then I'll re-run the suite and merge.

[GTC2080]

Rebased #2836 onto current main-v2 and pushed b7658c09.

What changed in the conflict resolution:

• Folded the current final-readiness gate into finalReadinessCheck.
• Preserved fix(agent): block final answers with incomplete todos #2920/fix(agent): allow plan-mode todos in final readiness #2933 behavior: incomplete latest todos block final answers outside plan mode, while plan mode todos remain allowed.
• Added audit counting for incomplete todos via readiness_incomplete_todos.
• Kept the new compactions metric in run --metrics while retaining readiness counters.

Local verification:

• D:\Go\go\bin\go.exe test ./internal/evidence ./internal/event ./internal/agent ./internal/cli
• D:\Go\go\bin\go.exe test ./internal/...
• D:\Go\go\bin\go.exe test ./...
• git diff --check origin/main-v2...HEAD

Note: the first full local run without env isolation picked up my user-global config and failed boot tests on an unconfigured planner_model. I reran the same commands with a temporary HOME/USERPROFILE/XDG_CONFIG_HOME/AppData, matching a clean CI-style environment, and they passed.