feat: stamp node_version and node_commit on every json log line

[jacderida]

Linear issue

Closes V2-1153

Risk tier

• T0 — docs / tooling / CI / pure UX-output. Repo CI only.
• T1 — client-only, no network-facing behavior change. CI + prod compat smoke.
• T2 — node/client logic with behavioral surface, no protocol/format/economics change. Dev testnet + ADR.
• T3 — protocol / storage format / payments / routing. T2 evidence + adversarial testing.

Proposed T1: the only behavioural change is two extra fields on each JSON log line. No wire, storage, payment, routing or upgrade-mechanism change.

Compatibility

• Wire: none
• Storage: none
• API: none. JSON log lines gain node_version and node_commit fields; the text format and the startup line's existing version / commit fields (parsed by the ant-client beta forwarder) are unchanged.

Semver impact

• breaking
• feature
• fix

Summary

Every document in production Elasticsearch carries tag.node_version = "unknown", so the build a node is running cannot be read from telemetry. That made the 0.18.1 staged rollout unreadable: fleet-wide relay recv stream closed errors rose 6x across the window, which looked like a regression in the build that fixed them; only splitting by the build each service was actually running (inferred from process start time) showed the new build was ~4x better (V2-986).

The tag is a static Telegraf [global_tags] value set at provisioning time in saorsa-testnet-registry and never passed by any caller. Even if it were passed, a provisioning-time string can never be right across an auto-upgrade, so the process itself has to report its build.

• ant_node::logging::WithBuildInfo wraps the JSON event formatter: it renders the inner event into a thread-local buffer and replaces the closing brace with a compile-time ,"node_version":"<CARGO_PKG_VERSION>","node_commit":"<ANT_GIT_COMMIT>"} tail. One memcpy per line on top of the serialisation already done; re-entrancy falls back to a fresh String rather than panicking; non-object output passes through untouched.
• Why a wrapper: tracing-subscriber's JSON formatter has no hook for constant fields, and span fields do not survive tokio::spawn, so a root span would miss most lines.
• Both JSON sinks (stdout and daily-rolling file) use it via json_event_format() in main.rs.
• Auto-upgrade: the upgrade path either execs the new binary or exits for the service manager to restart it, so the new process stamps its own constants — correct on both sides of the restart by construction.

Companion change: jacderida/saorsa-testnet-registry adds node_version, node_commit to Telegraf tag_keys so the fields land as tag.node_version.keyword / tag.node_commit.keyword and beat the static global tag. Until that config is re-provisioned on ant-prod-01, the values still land as top-level node_version / node_commit fields, so they are queryable either way. The bootstrap-upgrade-check / track-upgrades-setup skill notes saying tag.node_version is static remain true until both have rolled out.

Test evidence

• 3 new unit tests in src/logging.rs: every line is valid JSON and keeps message/level/target/custom fields; the stamp lands after a nested "span":{...} object; node_version equals CARGO_PKG_VERSION and parses as semver.
• cargo test: 1088 lib + 16 integration passed, 0 failed.
• Smoke run in development mode: stdout sink 295/295 lines and file sink 81/81 lines carried node_version=0.18.1 node_commit=31fcbae, 0 invalid JSON (jq-checked); text format byte-identical.
• cargo clippy --all-features -- -D clippy::panic -D clippy::unwrap_used -D clippy::expect_used clean; CI's cargo clippy --all-targets --all-features -- -D warnings clean; cargo build --no-default-features compiles.

Sample line:

{"timestamp":"2026-09-11T23:58:07.002929Z","level":"INFO","message":"Bootstrap peers provided via CLI","count":1,"target":"ant_node","node_version":"0.18.1","node_commit":"31fcbae"}

New dependency

none

ADR

n/a

Mitigation / rollback

Revert this commit; the two fields disappear and Telegraf's static global tag takes over again. No stored data or peers are affected.

🤖 Generated with Claude Code

https://claude.ai/code/session_01MYvuVpFYXXNrY9YFVBswvp