feat(provider-dashboard): v2 corrective — preserve v1 top, replace events UI with 4 v2 zones

[AquiGorka]

Summary

Corrective v2 layout per redesign-provider-dashboard-3-prompt.md. Closed PR #33 (wrong layout — v1 replacement based on a misread of the sketch wording). This PR is fresh and:

Preserves v1 (code paths unchanged)

• Top header + copy URL / copy OpEx / Fund chips
• OpEx Balance card
• 3-up Councils list (asset-chip-copy dropped per -3 §5)

Removes the entire v1 events UI

• 5-column tx flow (Deposit / Mempool / Submitted / Verified / Withdrawn)
• renderTxDetail modal + renderUtxoDetail + renderWithdrawDetail
• Live/Range mode toggle, Range Search, date pickers, auto-flip-to-Range
• setupDashboard orchestration + classifyTxIntoColumns
• api.ts wrappers exclusively serving the removed surfaces: listTransactions, getTransactionDetail, getUtxos + their types

Adds 4 always-live v2 zones in the freed space

• Counter strip (4 boxes; #1971c2 outline / #e7f5ff fill): THROUGHPUT (last-15m mean of throughputPerMin), AVG LATENCY (weighted avg avgProcessingMs over the last 100 bundles), QUEUE DEPTH (most recent snapshot), ERROR RATE (1h bundlesFailed / (completed + failed + expired); gracefully — until PR-104 lands and exposes bundlesFailed).
• Topology: SVG edges + HTML nodes. MY PP yellow ellipse at center, council ring, sibling-PP dots via POST /dashboard/council/discover (cached per render, best-effort). Live edge pulses on every relevant WS event with per-kind color per sketch. Council node palette transitions idle → low → high based on rolling-5-min pulse count (≥10 high, ≥3 low). Passive — no click handlers anywhere.
• Activity feed (right rail): card stack, newest top, ≤5 visible, ~8s fade, border-color per event kind, deposit/withdraw amounts surfaced. Passive — no click handler.
• Sparklines (3 charts, blue/purple/green): rolling 60-min trends for throughput / latency / queue from /dashboard/metrics?rangeMin=60, re-fetched every 60s alongside the counter strip.

WS reconnect handled by EventsClient (existing exponential-backoff path). No Range fallback when WS drops, no mode toggle anywhere.

ERROR RATE counter gracefully degrades if the platform predates PR-104: when the snapshots don't include bundlesFailed, the counter shows —. Once PR-104 merges + deploys, the counter starts populating without any further frontend change. MetricsSnapshot.bundlesFailed is typed as optional for that reason.

Out of scope (per -3)

• No click-wiring on new zones (drill-down destinations are gone with the events UI; future click behavior is a separate change).
• No backend changes; no new endpoints. provider-platform#104 (bundles_failed) stays open as a separate PM-owned merge.
• No mobile / responsive redesign.
• No restoration of asset-chip-copy on the 3-up Councils list.

Test plan

• deno task check clean (type-check)
• deno fmt --check clean
• deno lint clean
• deno task build (dev + production) succeeds
• deno task test green (7 passed)
• PM verifies in browser per -3 §6:
  • v1 top section unchanged (header chips + OpEx card + 3-up Councils)
  • No 5-column / no tx detail card / no Live/Range / no Search visible
  • 4 v2 zones render below the v1 top section
  • Live WS events animate pulses + counter increments + feed cards + sparkline ticks
• provider-platform#104 separately verified end-to-end once merged + deployed; ERROR RATE counter then populates with real values.

Companion file: PROVIDER_DASHBOARD_NOTES.md not added — Phase 0 raw audit dumps from -1 still apply unchanged.