Add Status tab with auto-revival, latency-based smart routing, cost tracking, and bug fixes

[MIsmail3018]

What

Three new features for the rotato admin dashboard and key rotation pipeline:

1. Auto-Revival - re-tests quarantined keys every 1 hour against their provider base URL and auto-unquarantines them if they come back online. Prevents permanent key loss when a provider briefly rejects requests.
2. Smart Routing (latency-based) - rolling 20-sample latency window per key with avg/p50/p95 stats. getNextKey() now sorts candidates by average latency, fastest first. State persists across restarts.
3. Per-Model Cost Tracking - records tokens_in / tokens_out / cost per request from upstream usage blocks. 24h summary by provider AND by model. 10k-record ring buffer.

New REST endpoints (admin auth required)

• GET /admin/api/status - named keys + dead keys + 24h usage
• GET /admin/api/latency - latency stats per key
• GET /admin/api/cost - cost/token summary 24h + by provider/model
• GET /admin/api/quarantine - full quarantine state
• POST /admin/api/quarantine/clear - body {key?} clears one or all
• POST /admin/api/revival/run - triggers AutoRevival immediately

Dashboard

New "Status" tab in public/admin.html with:

• 4 stat cards (providers, named keys, dead keys, requests/24h)
• Dead keys panel with human names + quarantine expiry
• Per-provider quota bars (red >=90%, yellow >=70%, green <70%)
• Per-key usage rows with 429 count
• Smart Routing card (blue left-border, latency rank with medals)
• Cost Summary card (green left-border, 24h totals)
• Auto-refresh every 10s when tab active

Files changed

• public/admin.html - +279 lines (Status tab + 3 new render functions)
• src/keyRotator.js - +366 lines (LatencyTracker, CostTracker, AutoRevival classes)
• src/openaiClient.js - +42 lines (latency recording on success/error, cost recording)
• src/server.js - +205 lines (6 new endpoints, AutoRevival wiring)

Total: 4 files changed, 858 insertions(+), 34 deletions(-)

Tested

2026-06-07 with 14 real requests: 6 keys tracked, latency range 1053ms-4601ms, 13497 tokens, 0 errors. AutoRevival scheduler fires at 30s after startup and every 1h thereafter.

Notes for reviewer

• AutoRevival uses the same default base URLs as the openai client; if a provider has a custom BASE_URL in .env, it will be picked up from config.providers.
• Quota limits in the dashboard (50/day OpenRouter, 300/day OmniRoute, etc.) are hardcoded in JS - they should be made provider-configurable in a follow-up.
• The cost field is recorded as 0 because rotato does not currently know per-model pricing; a follow-up could parse model_pricing.json if you maintain one upstream.

---

Update: Bug fixes from user review

After the first review surfaced 4 data bugs in the Status tab, this branch now also includes:

1. Providers stat — was showing 0 (Object.keys on Map). Fixed with Array.from(getProviders().keys()).
2. omni-route-key dead — was marked dead from HTTP 403, but 403 means endpoint unavailable, not key invalid. Removed 403 from deadKeyStatusCodes; AutoRevival.testKey now distinguishes true=alive, false=key-dead, null=endpoint-unavailable.
3. Quota over-count — was including rotato's own auth-gate rejects. Filtered at server, exposed as separate auth_rejected_24h stat.
4. (no key) rows — was polluting per-provider display. Hidden in dashboard, replaced with new Auth Rejected stat card.

Also filtered out zero-count ghost providers (e.g. 'openai'/'unknown' with 0 requests) from the response.

Total: 4 files, +40/-11 on top of the original +858/-34.