infra(railway): migrate health-prober + rollups off CF isolates

[JSONbored]

Move the */15 health probe and the hourly/daily rollups to Railway (worker/cron) against Postgres; use Timescale continuous aggregates for the heavy windows so the Worker reads pre-aggregated tables. Removes the isolate row-cap gymnastics. Deliverable: prober + rollups on Railway, CF crons retired, stable-IP probing.

Part of #2108. See ADR 0013 (docs/adr/0013-hybrid-deployment-topology.md) + runbook (deploy/README.md).

[JSONbored]

Correcting the target for this issue: the original plan (move health-prober + rollups to Railway) is stale. Railway's Postgres/Redis/indexer-rs stack was fully retired in #2956 (closed) — the self-hosted bare-metal box (`meta-indexer-01-us-lax1`) is now the permanent core per ADR 0014, which supersedes ADR 0013 that this issue originally referenced. Railway's only remaining role anywhere in this system is `wss-lb`.

The underlying goal is still valid and still open: move the `*/15` health probe and the hourly/daily rollups off Cloudflare Worker isolates (removing the row-cap gymnastics D1 forced) onto the self-hosted box against Postgres/TimescaleDB, using Timescale continuous aggregates for the heavy windows so the Worker reads pre-aggregated tables instead of computing them per-request.

Not yet actionable this session — this is a genuine new-infrastructure build (a cron/worker process on the box, TimescaleDB continuous aggregate definitions, a migration path for the existing CF-cron-driven rollups in `workers/api.mjs`), not a cleanup of already-shipped work. Leaving open, retargeted to the self-hosted box instead of Railway.