feat(eth2api): add Charon-parity beacon node request metrics

[varex83]

Summary

Adds Charon eth2wrap-parity Prometheus metrics for beacon node requests, using Pluto's vise conventions (not the prometheus crate, which Pluto doesn't use):

• app_eth2_requests_total{endpoint}
• app_eth2_errors_total{endpoint}
• app_eth2_latency_seconds{endpoint} — with Charon's exact 17 latency buckets

Metric names match Charon's app_eth2_* series. They're defined in a new crates/eth2api/src/metrics.rs and auto-registered via #[vise::register], so the existing /metrics endpoint surfaces them with no extra wiring.

A small instrument(endpoint, fut) helper reproduces Charon's per-call semantics — record a request + observe latency always, increment errors on Err — and slots in ahead of the existing .map_err(...) handling at each call site (the generated client methods return anyhow::Result, so caller types are preserved).

Coverage

Instruments the endpoints Pluto actually calls today, using Charon's snake_case labels:

• Inside existing eth2api wrappers (zero call-site churn): attester_duties, all submit_* duties, spec, genesis, fork_schedule, validators.
• Inline at core/app call sites: proposal, attestation_data, sync_committee_contribution, aggregate_attestation, proposer_duties, sync_committee_duties, node_syncing, node_version, node_peer_count.

Endpoints Charon instruments but Pluto doesn't call yet (e.g. beacon_block_root, the *_selections, prepare_sync_committee_subnets) are intentionally skipped — each is now a one-line instrument(...) addition when the port reaches them.

Deliberate deviations from Charon

1. using_fallback gauge omitted — Pluto has no multi/fallback beacon client yet (Implement app/app.go equivalent — node graph wiring & lifecycle #402 part B). To be added when fallback lands.
2. errors_total counts transport/decode Err only, not HTTP 4xx/5xx. Charon's go-eth2-client returns an error on non-2xx; Pluto's generated client returns Ok(Response::BadRequest(..)) for 4xx, which wrappers convert to Err after the instrumented call. Full error-count parity would require instrumenting around response classification — can follow up if exact parity is wanted.

Testing

• cargo +nightly fmt --all --check ✅
• cargo clippy -p pluto-eth2api -p pluto-core -p pluto-app --all-targets --all-features ✅ (clean)
• cargo test: eth2api (119) ✅, core (600) ✅, app ✅ — includes new unit tests asserting the exposition text and all latency buckets.

🤖 Generated with Claude Code