fix(jury): adjudication consent gate + reviewer no-show pause

[t-bhendarkar]

Summary

Implements the two documented-but-unenforced eligibility gates found during the user-journey doc fact-checks (issues registry #71/#72). Stacked on #97 (includes its commit until it merges).

Commit 1 — adjudication consent gate (selectJury / selectExperts)

Both journey docs (JUROR.md, EXPERT.md) describe the "I want to help adjudicate" toggle as gating panel seats — "same toggle as Juror." In code, reviewer_consent was enforced only for reviewer selection; jurors and experts were drafted from every personal identity above the score floor, then charged the −1 no-commit penalty for ignoring summonses they never agreed to receive.

• Same consent contract as reviewer-selection (true or legacy 1)
• Hard filter — never relaxed by the score/geo cascade; an all-non-consenting pool returns insufficient: true rather than drafting anyone
• Tests: non-consenting identities never seated (jury + experts), numeric consent accepted, insufficient-pool behavior

Commit 2 — reviewer no-show pause

REVIEWER.md claims "your no-show count goes up and can pause your assignments" — no counter or consequence existed; a reviewer could ignore every assignment forever, burning a 48h SLA per flagged content.

• getReviewerNoShowCount: pure read over DAG history — RECUSED reviews with decision_note = sla_expired (node-signed auto-recusals) within the last NOSHOW_SAMPLE_SIZE resolved assignments. Manual recusals and still-open assignments never count. No new state.
• Gate: more than MAX_NOSHOW_RECUSALS (3) within the window (10) ⇒ paused. Hard filter — relaxing would re-assign the case to someone already ignoring assignments.
• The "sla_expired" literal (written by the trigger, now matched by selection) is promoted to RECUSAL_REASONS.SLA_EXPIRED in shared/constants so the two sites can't drift.
• New genesis constants max_noshow_recusals / noshow_sample_size (selection must agree across nodes) ⇒ genesis payload changed ⇒ npm run seed re-run regenerated GENESIS_TX_SIGNATURE / GENESIS_VP_TX_SIGNATURE and genesis-data artifacts per the documented procedure.
• Tests: only sla-expired counts, threshold boundary, rolling-window recovery by serving, open assignments don't consume the window, paused reviewer excluded at every cascade pass.

⚠️ Consensus-affecting — coordinated upgrade

Selection is deterministically replayed by every node; nodes on different versions would seat different panels for the same dispute. Additionally the genesis-constants change means a new genesis hash — fine for dev clusters (reset by convention), but this and #97 should deploy in one coordinated window.

Operational note: from this version, adjudication requires opt-in — pilot/seed users must have reviewer_consent set or jury selection will (correctly) return insufficient.

Test plan

• New: tests/jury-consent.test.js (4) + no-show describe in tests/reviewer-selection.test.js (5)
• Full suite: 1954 passed / 1996, 0 failures on final run (one intermediate run hit the known pre-existing roaming flake — rotation-coordinator this time — which reproduces on pristine main and is tracked in the issues registry)
• Genesis verification tests green after re-seed