feat(cli): extend --retry-errors to resume missing cases from crashed runs

[christso]

Closes #975

Summary

• Extends --retry-errors to also re-run test cases that are missing from the previous output (e.g., due to a crash or Ctrl+C mid-run), not just execution_error cases
• Uses a micromatch negation filter to exclude completed (non-error) test IDs, so any test not in the previous output runs naturally
• No new checkpoint system, CLI commands, or flags — reuses existing index.jsonl artifacts
• Matrix-safe: only excludes test IDs where ALL results across targets are non-error
• Escapes glob metacharacters in test IDs for correct literal matching

How it works

Previously --retry-errors built an inclusion filter matching only error test IDs. Now it builds an exclusion filter (!{completed-1,completed-2,...}) that skips already-completed tests. This naturally includes:

1. Tests that errored (execution_error)
2. Tests that never ran (missing from output due to crash)

Test plan

• Unit tests for loadFullyCompletedTestIds(), buildExclusionFilter() (12 tests pass)
• Build, typecheck, lint all pass
• Manual e2e: partial run (2/5 completed) → --retry-errors runs 3 missing cases, merges 2 preserved
• Manual e2e: all completed → prints "No execution errors or missing cases. Nothing to retry."
• Manual e2e: matrix case (same test ID ok on one target, errored on another) → re-runs on all targets (conservative)

🤖 Generated with Claude Code

[cloudflare-workers-and-pages]

Deploying agentv with    Cloudflare Pages

Latest commit:	9b392e8
Status:	✅  Deploy successful!
Preview URL:	https://83b6df70.agentv.pages.dev
Branch Preview URL:	https://feat-975-retry-resume-missin.agentv.pages.dev

View logs

[christso]

E2E Verification Results

Test 1: Crash recovery — missing cases are re-run

Red (before): --retry-errors on a partial output (2/5 tests completed) would print "No execution errors found. Nothing to retry." and exit, losing the 3 missing cases.

Green (after): --retry-errors on the same partial output:

Skipping 2 already-completed test(s).
0/3   🔄 code-quality-multi-eval | llm
0/3   🔄 summary-task | llm
0/3   🔄 summary-multi-criteria-score-ranges-proposed | llm
...
Merged 2 non-error result(s) from previous output.
Total tests: 5

Correctly ran the 3 missing cases and merged the 2 preserved results.

Test 2: Error cases still re-run (backward compat)

--retry-errors on output with execution_error cases:

Found 3 execution-error test(s): code-explanation-simple, technical-writing-detailed, code-quality-multi-eval
Skipping 1 already-completed test(s).
0/4   🔄 code-explanation-simple | llm
...
Merged 1 non-error result(s) from previous output.
Total tests: 5

Error cases re-run, completed cases preserved, total correct.

Test 3: All completed — nothing to retry

--retry-errors on output where all 5 tests passed:

Skipping 5 already-completed test(s).
All tests completed successfully in the previous run. Nothing to retry.

Exits cleanly without re-running anything.