central prometheus, strip per-harness ops

[Flotapponnier]

Refactors the harness model to a clean producer/scraper split: every harness exposes /metrics, a single shared Prometheus scrapes them all.

What's added

• infrastructure/prometheus/ - single shared Prom config + Dockerfile, scrape jobs for both harnesses
• infrastructure/README.md - documents the architecture, how to add a target, why no Grafana/Alertmanager

What's removed

• harnesses/aggregator-head-lag/ - monitoring/, prometheus/, grafana/, assets/, docker-compose.yml, Makefile, grafana-entrypoint.sh, cmd/script/update_metrics_calls.sh
• harnesses/bridge-monitor/deploy/ - whole folder (per-harness Prom + Grafana + Alertmanager + docker-compose)

What's updated

• Harness READMEs reflect the new contract (data producer only, no infra packaging)
• harnesses/README.md restated for the central model
• Root README.md architecture diagram + 'Running locally' section
• CONTRIBUTING.md adds step 4 (append scrape job to infrastructure/prometheus/prometheus.yml)

Net effect

• 9988 lines deleted (per-harness Grafana JSON dashboards, duplicate Prom configs)
• 265 lines added (clean central infra)
• Each harness folder is now ~5 files: cmd/, Dockerfile, .env.example, .gitignore, README

Follow-up (Railway)

After merge: create OpenChainBench Railway project with 3 services - prometheus (from infrastructure/prometheus/), aggregator-head-lag, bridge-monitor (each from harnesses//). Update YAML prom_url to the new Prom URL. Old Mobula Railway services can stay up as fallback during migration.