fix(registry): crawler probes all registered agent types, not just sales

[bokelley]

Closes #4213

Summary

startPeriodicCrawl was seeded once at startup with a static snapshot of sales-only agents. Signals, buying, and creative agents enrolled in the member registry were excluded from the periodic health + capability probe and stuck permanently on stale "degraded" status with "No health snapshot yet". Agents registered after server startup were also never picked up until restart.

Two root causes fixed:

1. Filter bug (server/src/http.ts): listAgents("sales") → listAgents() so all registered agent types enter refreshAgentSnapshots on every crawl cycle.
2. Static-list bug (server/src/crawler.ts): callback-getter pattern re-fetches the agent list on every tick so newly enrolled agents appear within one 6-hour cycle.

POST /api/crawler/run (admin on-demand trigger) updated from listAgents("sales") to listAgents() so manual and scheduled runs behave identically.

Non-breaking justification

Server-internal scheduler change only. No schema, task definition, or public API surface touched. The change is strictly additive — a larger probe set means more agents get health/capability snapshots written, not fewer. crawlAllAgents already filters non-sales agents out of the publisher adagents.json walk internally (if (agent.type !== "sales") continue), so passing all agent types is safe.

members_only agents remain intentionally excluded from the periodic crawl (viewerHasApiAccess defaults to false). Owner-triggered per-agent probes for members_only agents continue through POST /api/registry/agents/:url/refresh (which passes includeMembersOnly: true).

Pre-PR review

• code-reviewer: approved — both root causes correctly addressed; changeset, POST /api/crawler/run consistency, and members_only scope comment addressed before this draft opened
• internal-tools-strategist: approved — callback-getter pattern correct; performance watch item noted (probe count scales with registered agent count at CONCURRENCY=5 / 10s timeout; worth monitoring probed count in "Agent snapshots refreshed" log line post-deploy)

Nit from reviewers (noted, not fixed): startPeriodicCrawl logs 'Periodic crawl started' before the initial run() resolves — cosmetic ordering issue, no functional impact.

Triage-managed PR. This bot does not currently iterate on
review comments or PR conversation threads (only on the source
issue). To unblock:

• Push fixup commits directly: gh pr checkout <num> →
  fix → push.
• Or re-trigger: comment /triage execute on the source
  issue.

See #3121
for context.

Session: https://claude.ai/code/session_01Qm6rGjKvgavkV9rynMebXx

---

Generated by Claude Code