Export low-cardinality usage metrics for Explain Error

[shenxianpeng]

Problem

Administrators need observability into Explain Error adoption and cost-related behavior, but there are currently no aggregated metrics.

Goal

Expose low-cardinality usage metrics for monitoring systems.

Scope

Export aggregated counters/histograms for:

• total requests by entry point and result
• provider calls by provider/model/result
• request duration
• approximate input size
• quota rejections (for future compatibility)

Examples:

• explain_error_requests_total{entrypoint,result}
• explain_error_provider_calls_total{provider,model,result}
• explain_error_request_duration_ms
• explain_error_input_log_lines

Non-Goals

• Exporting raw logs, prompts, explanations, or job/build identifiers
• Per-job metrics labels
• Audit UI
• Quota enforcement

Important Constraints

Metrics must not use high-cardinality labels such as:

• full job name
• build number
• build URL
• username

Proposed Implementation

• Add a metrics-backed UsageRecorder implementation.
• Wire it to the internal usage events introduced previously.
• Reuse Jenkins-friendly metrics mechanisms where possible instead of inventing a custom one-off format.
• Keep metrics export optional or safely inert when no exporter/integration is configured.

Acceptance Criteria

• Metrics reflect requests from both entry points.
• Cache hits are counted separately from real provider calls.
• Provider/model dimensions are available.
• Metrics contain no high-cardinality labels.