host.name .local suffix is not normalized on the Prometheus metrics path

[ryanrozich]

Context

Raised by a Codex review on catalyst-otel#148 (P2), where the proposed fix was reverted as inert rather than shipped.

transform/metrics_normalize copies resource.attributes["host.name"] into a host_name log attribute, but it contains only log_statements — it runs on the logs pipeline. Native claude_code_* metrics reach Prometheus through a different path (resource_to_telemetry_conversion on the prometheus exporter), which promotes the raw resource attribute untouched.

Consequence: if any emitter ever supplies host.name=foo.local, Prometheus gets host_name="foo.local" while the logs path gets the .local stripped — one machine split across two series, which is the exact failure CTL-812 fixed for the hostname label.

Why this is filed rather than fixed

Nothing has been observed emitting a .local host.name to this stack. Every current emitter sets a canonical short name (laptop, mini, mini-2) explicitly. Adding a second speculative OTTL statement to a 73k-line config to guard an unobserved case is how that file got hard to reason about; a measured occurrence should drive the fix.

A .local value was observed once — in a deliberate synthetic tunnel probe (host.name=container-probe.local), which confirms the promotion path is unguarded but is not evidence of a real emitter.

Acceptance criteria

Given an emitter sends host.name=<name>.local on native claude_code_* metrics
When those metrics reach Prometheus
Then the host_name label reads <name>, matching the logs path

Given the fix is in place
When an emitter sends an already-canonical host.name (e.g. laptop)
Then the label is unchanged (the statement is a no-op)

Notes

Likely shape: a metric_statements block in a metrics-pipeline transform applying the same replace_pattern(..., "\\.local$", "") that transform/logs uses for hostname. Verify with otelcol-contrib validate plus a negative control (a deliberately broken statement must fail), since this repo has no CI.