Adopt the shared Milo instrumentation library #1144

Goal

Adopt the shared Milo instrumentation library (milo-os/instrumentation) in every Galactic binary, so Galactic's metrics and logs follow the same conventions as other Milo services.

The library standardizes three things: a production logger (structured JSON at info level, carrying the binary's version and the library's version on every line), a single metrics registry whose constructors refuse label names the collection pipeline reserves, and a build-info metric to join on when breaking a query down by version. A conformance test fails the build when any of these stops being true. The adoption guide is ADOPTING.md.

Galactic is already close on metrics: nothing registers on Prometheus's default registry, and most metrics are served from the controller-runtime registry. The main gaps are:

• Several long-running binaries hand controller-runtime a development-mode logger in production, so its log lines are unstructured console text at debug level.
• Logging is built differently in each binary (slog JSON, slog text, zap, or a slog bridge).
• Metric constructors receive a registry as a parameter, and a few components serve metrics from a private registry. The library expects one shared registry.
• A few metrics use label names the collection pipeline reserves for itself (node, namespace, provider, component). These collide with the pipeline's own values, and some alert rules group by them.
• Many datapath metrics are custom collectors that compute values at scrape time. The library registers these but can't check their label names, so they need a manual review.

The library requires Kubernetes 0.37 libraries and controller-runtime 0.25 or later. Galactic is on earlier versions, so that upgrade comes first.

Done when

• Every long-running Galactic binary logs through the library's logger in production format. The CNI plugins write the same format to their log file. Existing log call sites keep working through a slog bridge.
• Every binary reports its version through the library, and the build-info metric is served beside its other metrics.
• Every metric is created through the library or registered through it as a custom collector, on the one registry each binary serves. No component serves a private registry.
• No metric sets a label the collection pipeline reserves. Each alert rule, its rule tests, and every downstream copy of the rules match the renamed labels before the rename ships.
• The library's adoption test runs in CI with no skipped directories, alongside its metric-name lint and a scrape test for each binary's metrics endpoint.
• The repository's agent instructions state the rule, so coding agents follow it before the test fails.

Out of scope

• New metrics, or changes to what existing metrics measure.
• Tracing.
• Changes to the collection pipeline itself.

Milestones

• Use a production-format logger for controller-runtime in every binary #1145
• Upgrade to Kubernetes 0.37 libraries and controller-runtime 0.25 #1146
• Adopt the standard logger and build version in every binary #1147
• Add the instrumentation adoption test with a shrinking skip list #1148
• Create every metric through the instrumentation library #1149
• Rename metric labels reserved by the collection pipeline #1150
• Require the instrumentation library in Galactic's agent instructions #1153