Firehose phase 2: follows hose (kind 3) + relay-URL harvesting

[melvincarvalho]

Phase 2 of the firehose rework (after #7). Adds the follows hose (kind 3) and retires the legacy raw-upsert path for kind 3. First real exercise of the HOSES switch (HOSES=profiles,follows).

Background

The follows collection uses a derived shape — { pubkey, follows: [hex…], created_at, count } — not raw events. That matters: the in-degree query (followerCount → countDocuments({ follows: pubkey })) and enrichCounts both depend on the top-level follows array + count. The legacy upsertEvent fallback (current phase-1 path for kind 3) writes the raw event instead, which has no follows array — so follower counts silently don't accrue for events it writes. Moving kind 3 to a proper hose fixes that.

The old followshose.js also harvested relay URLs from the kind-3 content (legacy NIP-65 relay map, {"wss://…":{read,write}}) into the relays collection. We keep that discovery side-effect.

Scope

• src/hoses/event.js: extract the shared structural + schnorr-signature check (eventId, verifySignature) used by every hose; refactor the profiles hose to use it (DRY).
• src/hoses/follows.js (kind 3):
  • Verify the event signature (kind 3 content is optional — unlike kind 0, it is not required to be JSON).
  • Derive { pubkey, follows: [hex…], created_at, count } from p tags (validate 64-hex, dedupe).
  • Latest-wins upsert (the old followshose blindly replaced — an out-of-order older event could clobber a newer list; we gate on created_at).
  • Owns its indexes: pubkey, follows (reverse lookup), created_at.
  • Relay harvesting: parse content as a relay map, canonicalize URLs (trailing slash on origins; ws/wss only), $setOnInsert into the relays directory so discovery never overwrites health metrics. Best-effort — never fails ingest.
• src/indexer.js: register the follows hose. The switch then removes kind 3 from the legacy set automatically; HOSES=profiles,follows becomes a meaningful selection.
• Tests: verify/parse/harvest unit tests; planIngest two-hose cases already exist from Firehose phase 1: validated profiles hose (kind 0) + extensible hose seam #7.

Out of scope (later phases)

Relay-lists hose (kind 10002) · relay-health prober · deploy/retire legacy server hoses.

Acceptance

• npm test green incl. new follows-hose tests.
• Default indexer now routes kind 3 → follows hose (derived shape, in-degree works); kind 10002 still legacy until phase 3.
• Forged/tampered kind-3 events rejected; relay URLs from content discovered into relays without clobbering existing health docs.