split 4 harnesses: 1 bench = 1 service

[Flotapponnier]

Splits the two combined Go projects into four independent harness folders, one per benchmark, matching the '1 bench = 1 service' principle.

What changed

• harnesses/aggregator-head-lag/ — slim, only runs head_lag goroutine
• harnesses/metadata-coverage/ — new folder, only Pulse feeder + metadata worker
• harnesses/bridge-quote-latency/ — clone of bridge-monitor, dedicated for bench №002
• harnesses/bridge-fee/ — clone of bridge-monitor, dedicated for bench №003
• Old harnesses/bridge-monitor/ removed

YAML source paths updated

Each spec's source: now points at its own harness folder.

Prometheus scrape config rewritten

infrastructure/prometheus/prometheus.yml now has 4 jobs (one per benchmark) instead of 2. Target hosts use the convention -production.up.railway.app.

Follow-up (Railway, by Mobula)

• The existing aggregator-latency-benchmark Railway service stays as is, but now should be repointed at root harnesses/aggregator-head-lag (or its mobula-api/miniapps/aggregator-head-lag equivalent on dev branch)
• 3 NEW Railway services to create from this repo: metadata-coverage, bridge-quote-latency, bridge-fee
• Old bridging-latency Railway service can be retired once the 2 new bridge services are healthy