Prometheus metrics + pprof endpoints

## Summary
Expose Prometheus metrics and pprof to operate and profile the crawler.

## Motivation
- Visibility into throughput, latency, errors
- Ability to capture CPU/heap profiles

## Scope
- `/metrics`:
- `gotor_requests_total{code,host}`
- `gotor_fetch_latency_seconds` histogram
- `gotor_bytes_total`
- `gotor_queue_depth{host}`
- `/debug/pprof/*` behind a flag

## Acceptance Criteria
- Metrics scrapeable by Prometheus
- Short crawl shows non-zero counters and sane histograms
- pprof captures are possible during a run

## Tasks
- [ ] `internal/metrics` package
- [ ] Register metrics in server
- [ ] Flag-gate pprof endpoints
- [ ] README: how to profile