feat(services): dense services list — health, Δ7d baseline, issues, last-deploy

[Makisuo]

What

Redesigns the /services list to the approved "dense table+" direction (designed in Paper first): each row now carries a health signal, a P95 delta vs the trailing-7d baseline, an open-issues count, and a message-first Last deploy cell — replacing the raw commit-sha column.

How (data stays fleet-level — no N+1)

The page still issues a fixed number of fleet-level requests (verified in the network pane: overview, sparklines, facets, anomalies, baseline, issue counts):

• Health — new useServiceHealthSummary hook derives per-(service, env) health from the already-fetched alert incidents + anomalies; shared by the table and the sidebar's new client-side Health facet, so counts and badges can't disagree. Shown as a quiet dot next to the name (degraded/unhealthy only) + a destructive left rail, not a dedicated column.
• P95 · Δ7d — wires the existing fleet-level, hour-cached serviceHealthBaseline query (previously zero consumers); joined via the canonical baselineKey.
• Issues — new GET /v2/error_issues/service_counts → ErrorsService.countOpenIssuesByService, one Postgres GROUP BY service_name over actionable, error-kind, non-archived issues. Name-scoped only (env scoping would require the warehouse fingerprint intersection; deliberately out of scope).
• Last deploy — serviceOverviewQuery gains per-commit firstSeen: min(Timestamp) on the same scan; the cell derives rollout state (progress bar + breakdown tooltip) and an errors ↑ since flag (newest commit ≥2× older-commit error rate, ≥0.5pp, ≥50 spans both sides) client-side. Commit messages resolve lazily via the existing deduped per-sha commitQueryAtom — the one accepted soft fan-out, bounded by distinct live shas.

Reviewer notes

• serviceOverview's cachedDirect policy is bumped to version 2 because rows gained firstSeen (stale cached rows would otherwise miss the field for one TTL window).
• ServiceOverviewResponse is schema-open (Record<string, Unknown>), so no domain change was needed for the overview; the v2 OpenAPI surface test pins the new service_counts path.
• Rows are extracted into a memoized ServiceRow; baseline/issues/messages are progressive enrichment (Result.isSuccess fallback) and never block first paint.
• service.version was considered and skipped — it isn't materialized in ClickHouse (only CommitSha is); would need an MV column first.

Testing

• bun run test green across all 24 packages (new: query-engine compile assertions incl. min(Timestamp) AS firstSeen, PGlite test for countOpenIssuesByService, updated v2 surface test); bun typecheck green.
• Verified live against the dev stack: fleet-level request set confirmed, health filter + facet + footer tally work, columns render progressively, layout checked at 1280/1600px.

🤖 Generated with Claude Code

---

^{Need help on this PR? Tag /codesmith with what you need. Autofix is disabled.}

[pullfrog]

Your Pullfrog Router balance is exhausted.

You have a card on file but auto-reload is disabled, so runs paused once your balance went past the overdraft buffer.

Top up balance → · Enable auto-reload →

  ^{｜ ⚠️ this action is pinned to a commit SHA, which freezes the cleanup step — switch to @v0 or keep the SHA fresh with Dependabot ｜ Rerun failed job ➔ ｜ View workflow run ｜ via Pullfrog ｜ Using Claude Opus ｜ 𝕏}