Add closed-thesis pre-filter gate with reason logging

Extract the closed-thesis prune logic from decage (e2cde0f) into a standalone gate module invoked before filter_score. Gate loads a frozen list of closed thesis IDs (SENTINEL, prior killed forecasters) from brain/closed_theses.json, embeds each candidate, and compares cosine similarity to closed-thesis embeddings. If max similarity > 0.82, mark candidate as gate_rejected with reason='closed_thesis:<id>' and skip filter_score + argument moves. Wire into pipeline.js between genesis and filter_score. Add gate result to engine.db gate_log table (already exists for resolution_gate).

Candidates retreading closed theses are rejected ~30s earlier in the pipeline, avoiding ~3 LLM calls each (filter_score + argument + verdict). Estimated 8-15% of current filter_rejected rows will be reclassified as closed_thesis_gate rejections. Daily LLM spend on rejected candidates drops measurably.

Run gate over the last 200 rejected hypotheses. If the gate's reject set agrees with decage's existing in-genesis prune decisions on <85% of cases (i.e. wide disagreement), the similarity threshold or closed-thesis list is miscalibrated and the gate is not a faithful extraction.

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