CI: record per-job interop retry-rate telemetry

[lance0]

Context

PR #185 made privileged M-series interop jobs more tolerant of transient containerlab failures by retrying each topology in .github/actions/run-interop-test. The action emits warning annotations when a retry succeeds, but there is no durable per-job summary that shows how often retries are being consumed across runs.

Expected direction

Add lightweight retry-rate telemetry for interop jobs so we can distinguish real stability improvements from flake masking. This should stay CI-local and low-risk: job summaries, uploaded JSON/Markdown artifacts, or workflow annotations are enough; no external metrics service is required for the first slice.

Acceptance criteria

• Each interop job records topology label, attempts used, final result, and whether a retry absorbed a transient failure.
• The workflow summary exposes retry counts in a way reviewers can see from the Actions UI.
• Failed jobs still surface the first failing attempt clearly enough for debugging.
• Telemetry does not change pass/fail semantics from PR ci: bounded-retry M-series interop jobs #185.
• Add docs or comments for how maintainers should interpret retry-rate trends.

Related

• PR ci: bounded-retry M-series interop jobs #185: bounded-retry M-series interop jobs
• .github/actions/run-interop-test/action.yml