feat(node): surface healthy provider counts on /debug/node_config

[haiyuechen-nearone]

Closes #3764. Stacked on #3849 (top of the stack), merge bottom-up:

1. feat(node): probe foreign-chain RPC providers on startup #3848 — startup probe + per-provider logs
2. feat(node): expose foreign-chain RPC provider health metric #3849 — per-chain configured / healthy Prometheus gauges
3. feat(node): surface healthy provider counts on /debug/node_config #3850 (this PR) — healthy counts on /debug/node_config

The startup probe publishes its per-chain healthy-provider counts over a watch channel; the web server overlays them onto /debug/node_config as foreign_chains_provider_health (empty until the probe completes, always present, mirroring foreign_chains_provider_counts).

[claude]

Pull request overview

Wires a tokio::sync::watch channel from run_startup_health_check to the /debug/node_config handler so per-chain healthy-provider counts published by the startup probe surface alongside the existing configured counts. New ProviderHealthSnapshot type (BTreeMap<String, i64> keyed by chain.label()) is overlaid at request time; the field is always present, defaulting to {} before/without a probe.

Changes:

• New ProviderHealthSnapshot type in foreign_chain_health.rs; run_startup_health_check / set_healthy_metric now accept a watch::Sender and publish alongside the Prometheus gauge.
• WebServerState gains a watch::Receiver<ProviderHealthSnapshot>; debug_node_config clones the base response and overlays the current snapshot.
• NodeConfigResponse gains a foreign_chains_provider_health field.
• run.rs and tests.rs create/plumb the watch channel; test callers updated.
• E2E cluster helper fetch_node_config added; startup foreign-chain-health E2E asserts the field content on probed and skipped nodes.

Reviewed changes

Per-file summary

File	Description
crates/node/src/foreign_chain_health.rs	Adds ProviderHealthSnapshot; threads a watch::Sender through the probe; updates unit tests to construct a channel and assert on the receiver.
crates/node/src/web.rs	Adds foreign_chains_provider_health field, receiver on WebServerState, overlay in debug_node_config; new/updated unit tests.
crates/node/src/run.rs	Creates the watch channel and passes sender to the probe / receiver to the web server.
crates/node/src/tests.rs	Wires a dummy watch receiver into the one-node test harness.
crates/e2e-tests/src/cluster.rs	New fetch_node_config helper.
crates/e2e-tests/tests/startup_foreign_chain_health.rs	Asserts the healthy overlay on /debug/node_config for probed and skipped nodes.

Findings

Non-blocking (nits, follow-ups, suggestions):

• crates/node/src/web.rs:103,124 — NodeConfigResponse stores foreign_chains_provider_health as a field, but the value set by From<ConfigFile> is always the empty default and is always overwritten at request time in debug_node_config (web.rs:189-192). This is a small design smell: the field exists only to satisfy Serialize. Consider either (a) constructing the response from the base struct + snapshot at handler time (no default field on NodeConfigResponse), or (b) documenting on the field that its stored value is unused and set per-request. Not a correctness issue, but future edits could easily be confused into believing the stored value flows to callers.
• crates/node/src/foreign_chain_health.rs:114 — let _ = health_publisher.send(counts); silently drops the SendError, which for watch::Sender only happens when all receivers have been dropped. That's benign here (web server holds the receiver for the process lifetime), but if that invariant ever changes this failure mode becomes invisible. A debug! on the discarded error would be cheap insurance without changing behavior.
• crates/node/src/foreign_chain_health.rs:81 — Golden-supplied probes on NetworkKind::Local don't publish a snapshot when nothing was probed (probe returned early via Skip variants). This is consistent with the existing "skipped chain → empty map on /debug/node_config" behavior asserted by the E2E test for nodes 1 and 2, so no action needed — but worth noting explicitly in the doc comment of run_startup_health_check that snapshots are only published after a probe actually runs (not on Skip*).
• crates/e2e-tests/src/cluster.rs:777 — No timeout on reqwest::get. The default reqwest client has no timeout; a hung web-server response would hang the test until the harness-level timeout fires. Consider using a reqwest::Client with an explicit timeout for consistency with other health/metric helpers in the file.

✅ Approved