[Observability] Add optional Prometheus-compatible metrics output

[Dyu20705]

Problem statement

A lightweight metrics endpoint or text export would connect RunSift to familiar observability workflows.

Implementation notes

Explore a Prometheus-compatible metrics format for summarized workflow reliability signals while keeping the MVP CLI-first.

Checklist

• Define which metrics are safe to expose
• Implement textfile or optional endpoint output
• Document labels and cardinality risks
• Add fixture-based example
• Keep feature optional

Acceptance criteria

• Prometheus-compatible output is documented
• Metric labels avoid high-cardinality pitfalls
• Feature does not block CLI usage
• Example output can be generated locally

Epic

#19