fix(pr-fix): separate human decisions from retry exhaustion (#1218) - #1221

Unexpected input(s) 'escalate-on-risk-flags', valid inputs are ['github-token', 'repo', 'pr-number', 'ai-base-url', 'ai-api-format', 'ai-model', 'ai-api-key', 'ai-max-tokens', 'ai-temperature', 'ai-response-format', 'ai-tokens-param', 'anthropic-version', 'ai-fallback-base-url', 'ai-fallback-api-format', 'ai-fallback-model', 'ai-fallback-api-key', 'ai-primary-retries', 'ai-primary-retry-delay-sec', 'on-model-failure', 'fail-on-request-changes', 'fail-on-degraded-review', 'verdict-policy', 'non-blocking-finding-categories', 'inline-findings', 'inline-findings-max', 'validate-required-checks', 'required-check-validation-mode', 'review-routing-mode', 'ai-primary-model', 'ai-primary-base-url', 'ai-primary-api-format', 'ai-primary-api-key', 'ai-smart-model', 'ai-smart-base-url', 'ai-smart-api-format', 'ai-smart-api-key', 'ai-specialist-model', 'ai-specialist-base-url', 'ai-specialist-api-format', 'ai-specialist-api-key', 'ai-specialist-correctness-model', 'ai-specialist-security-model', 'ai-specialist-tests-model', 'ai-stream', 'ai-fallback-stream', 'allowed-source-hosts', 'related-code-context', 'related-code-max-bytes', 'harness-obligations', 'equivalent-paths', 'equivalent-paths-max-bytes', 'requirement-trace', 'linear-api-key', 'linear-issue-prefixes', 'linear-issue-timeout-sec', 'linear-enable-for-forks', 'system-prompt', 'system-prompt-file', 'system-prompt-mode', 'review-verbosity', 'standards-file', 'standards-file-candidates', 'publish-review-comment', 'publish-mode', 'allow-approve', 'allow-repo-policy-overrides', 'approve-forks', 'cleanup-previous-native-reviews', 'upstream-link-mode', 'context-limit-mode', 'model-context-tokens', 'primary-model-context-tokens', 'smart-model-context-tokens', 'fallback-model-context-tokens', 'primary-request-shape', 'smart-request-shape', 'repo-map-context', 'repo-map-max-bytes', 'pr-thread-context', 'pr-thread-max-bytes', 'review-threads-context', 'review-threads-max-bytes', 'deep-review', 'deep-review-timeout-sec', 'deep-review-max-tokens', 'deep-review-corpus-max-bytes', 'claim-falsification', 'enrichment-budget-sec', 'image-digest-budget-sec', 'evidence-providers-file', 'sarif-files', 'sarif-max-findings', 'evidence-provider-timeout-sec', 'evidence-provider-max-output-bytes', 'evidence-provider-parallelism', 'evidence-blocker-enforcement', 'evidence-enable-for-forks', 'tool-mode', 'tool-loop-wall-clock-sec', 'tool-loop-summarize', 'tool-loop-summarize-max-tokens', 'tool-max-requests', 'primary-tool-max-requests', 'smart-tool-max-requests', 'tool-max-rounds', 'tool-turn-timeout-sec', 'tool-corpus-max-bytes', 'tool-max-tokens-per-turn', 'tool-max-response-bytes', 'tool-allowed-gh-api-repos', 'tool-request-timeout-sec', 'search-url', 'tool-max-search-results', 'tool-failure-enforcement', 'tool-min-successful-requests', 'tool-enable-for-forks', 'tool-mcp-servers', 'tool-mcp-token', 'tool-mcp-name-prefixes', 'ai-request-timeout-sec', 'ai-connect-timeout-sec', 'ai-fallback-request-timeout-sec', 'ai-fallback-connect-timeout-sec', 'platform', 'forgejo-api-url', 'forgejo-token', 'forgejo-auth-method', 'forgejo-authorized-integration-audience', 'forgejo-skip-permission-preflight', 'skip-if-diff-unchanged', 'force-review', 'rereview-label', 'rereview-command', 'comment-marker', 'ci-status-check', 'ci-timeout-sec', 'ci-interval-sec', 'ci-skip-on-timeout']