docs: v23-decision-first — update public skill docs to new response contract

[deluonchain]

Public skill docs → v23-decision-first

Brings the installable deluskill docs in line with the deployed v23 oracle (delucognition #4, merged). Installing agents now get the correct response contract.

Changed files

• SKILL.md — version bumped, decision-header section added up top, comparables removed, pool ranking + 1h→15m→5m ladder + live priors documented, ?verbose=true gating, response-schema summary rewritten, example response replaced with a real v23 default-shape payload (BNKR, WATCH/hold), quant weights table re-anchored to fixed structure weight.
• README.md — leads with the decision header, comparables dropped from "what you get", verbose + provenance fields noted.
• references/response-schema.md — decision block documented first, comparables removed, selected_timeframe/candle_count/pool_source added, ?verbose=true section, pool/ladder explainer.
• references/oracle-response.schema.json — JSON Schema rewritten: decision required, comparables gone, new provenance fields, oracle_version const v23-decision-first, observed/summary documented as verbose-only.
• references/external-clients.md — TS/Python recipes read decision first with a sample agent gate; legacy ?chain=base query param dropped (Base is implicit).

Unchanged

Single CA, GET, $0.25, the optional ?social=true checkr two-step flow.

One flag for review

The example response in SKILL.md uses representative numbers (BNKR, hold/WATCH). The shape is exact; the values are illustrative, not a live oracle call. Skim it if you want the example to reflect a current pull.

No deploy here — this is docs only. The handler is already on main in delucognition.