feat: add operational counters for sampling, requirements, and tools (#467)

[ajbozarth]

Misc PR

Type of PR

• Bug Fix
• New Feature
• Documentation
• Other

Description

• Link to Issue: Fixes Add counters for operational metrics specific to Mellea's sampling and validation loops #467

Adds six new OpenTelemetry counters as part of the telemetry epic (#443),
giving operators visibility into retry behaviour, validation failure rates,
and tool call health alongside the existing LLM-level metrics:

• mellea.sampling.attempts / mellea.sampling.successes / mellea.sampling.failures — tagged by strategy class name
• mellea.requirement.checks / mellea.requirement.failures — tagged by requirement class name and failure reason
• mellea.tool.calls — tagged by tool name and status ("success" / "failure")

Follows the established plugin pattern: lazy-init counter globals + record_*
helpers in metrics.py, consumed by three new Plugin classes in
metrics_plugins.py (SamplingMetricsPlugin, RequirementMetricsPlugin,
ToolMetricsPlugin) that hook into existing hook events in FIRE_AND_FORGET
mode. No metric calls in business-logic files.

Also extends SamplingIterationPayload and SamplingLoopEndPayload with a
strategy_name field (backward-compatible, defaults to "") so plugins can
tag counters with the strategy class name.

Testing

• Tests added to the respective file if code was changed
• New code has 100% coverage if code as added
• Ensure existing tests and github automation passes (a maintainer will kick off the github automation when the rest of the PR is populated)

Attribution

• AI coding assistants used

[github-actions]

The PR description has been updated. Please fill out the template for your PR to be reviewed.

[ajbozarth]

This is parallel work to #882 which adds the other remaining metrics telemetry and whichever merges second will need a manual merge to deal with the overlapping git diffs.

[akihikokuroda]

This looks good to me as the implementation of the issue #467. I wonder how these counters are useful. They are grouped by strategies and requirements. They may not give useful insight without some more fine groupings.

[akihikokuroda]

LGTM. I have a general comment about the issue that this PR implements.

[ajbozarth]

This looks good to me as the implementation of the issue #467. I wonder how these counters are useful. They are grouped by strategies and requirements. They may not give useful insight without some more fine groupings.

@jakelorocco @psschwei I actually wondered about this while implementing. I thought the idea was interesting and valuable, but my work did not create an example of how it could be used (which maybe I should add?)