Add Commander override JSONL extractor as structured Corpus E feed for meta_engine genesis

Create meta_engine/scripts/extract_commander_overrides.js. It opens engine.db read-only (via openSourceDbReadonly) and runs: `SELECT h.id, h.title, m.output, m.created_at FROM moves m JOIN hypotheses h ON h.id=m.hypothesis_id WHERE m.move_type IN ('commander_kill','commander_override','commander_defer') ORDER BY m.created_at DESC LIMIT 20`. For each row, emit: {source_corpus:'E', source:'commander_override', title:`Commander ${decision_type}: ${h.title}`, text:`Commander decision: ${decision_type}. Reason: ${parsed_reason}. Phase at override: ${h.phase}. Archetype tags: ${h.archetype_id || 'none'}.`}. Write JSONL to meta_engine/corpora/commander_overrides.jsonl (truncate-and-rewrite each run). In meta_engine/cycle.js loadCorpus() function, after loading the existing Corpus E items, append any lines from commander_overrides.jsonl if the file exists (if absent, log '[corpus] no commander_overrides.jsonl — run extract_commander_overrides.js' and skip). Add --no-override CLI flag to cycle.js to skip this step (needed for falsifier A/B test).

The next meta genesis run after deploy will include the 3 recent Commander KILLs (ClaimGate, AI Tool Claim Verification, AI Control Failure) and 1 DEFER (Upgrade Gate) as explicit Corpus E items. At least 1 proposal in that run should cite 'commander_override' in its evidence field and explicitly connect the override pattern to a proposed engine change. Currently, the 2026-05-21 cycle loaded 4 Corpus E items and 0 Commander override items — the override signal is invisible to genesis.

Run `node meta_engine/cycle.js --dry-run` twice: once with --no-override, once without. Compare the evidence[] arrays across the two run outputs. If no proposal cites 'commander_override' as source in either run, the override JSONL is not influencing proposer output — the item text needs richer context (full override reason + rejected hypothesis excerpt). If citation appears only in the non-override run, the corpus is working as designed.

Rejected by filter_score. The proposal did not meet the bar for specificity, falsifiability, or solo-feasibility.