Add observed-buyer pre-check at start of argument.js to skip no-market debates

Add a second pre-check function observedBuyerCheck(hypothesis) at the start of argument.js (after the describability gate if both proposals are implemented; as a standalone gate if only this one is). Single Gemini Flash call using the hypothesis description plus any evidence already stored from evidence_search results. Prompt: 'Based only on the evidence provided and your knowledge of the market, is there observable proof that someone currently pays for this workflow or a close analog? Observable proof: named products with public pricing pages, job postings with explicit budget authority for this function, analyst reports citing existing category spend, or documented customer spend on adjacent workflows. Answer OBSERVED or NOT_OBSERVED with one evidence sentence citing the source.' Condition: if NOT_OBSERVED, set kill_reason = 'no_observed_buyer' and skip the full argument+council_verdict pipeline. This gate is distinct from filter_score's acute_pain_not_candy (which measures pain intensity) — this gate specifically tests whether money is already flowing in the category. Cost: ~$0.001 per hypothesis.

Of 9 recent deep_council_verdict runs, at least 3 of the 5 killed hypotheses would be caught here: d3786b (Agronomy Advisory — no observable paying buyer for AI-powered agronomy ledgers), c27754 (Medical-Device SME buying AI components — buyer leverage unverified), cc72cd (Bot-Promise Slip — 'no observed buyer' explicitly in verdict). At current volume saves approximately $0.16-0.25/week in deep move costs (council_verdict + competitor + distribution + 6 others at combined ~$0.12-0.18/hypothesis). The two non-convergent verdicts (7199a9 Recruiter Fit-Claim, 2ca131 Sprint Estimate Stress-Test) involve unverified buyers and might also have been caught.

Apply gate retroactively to 9 recent council verdicts using stored evidence. Gate must NOT kill hyp-2026-05-10-26fc18 (Copilot Promise Ledger — 'real trust gap and clean AE-fit' verdict, which implies observable buyer-adjacent category exists) or hyp-2026-05-14-5d7cca (Recruiter Fit-Claim Scorecard — 'real pain, open quadrant' verdict). Gate must kill d3786b (Agronomy Advisory) and c27754 (Medical-Device SME). If the gate kills any hypothesis that received 'real pain' language in its verdict, the observed-buyer threshold is set too high (confusing niche markets with non-markets).

Deferred. Engine logic for this proposal is sound or partial, but the build was not prioritised this cycle.