Throttle failed MX performance requests to the solved interval

Description

GitHub issue #785 reports that getDeviceAppliancePerformance raises a 400 Feature not supported for an MX without score support. In src/meraki_dashboard_exporter/collectors/devices/mx.py the per-serial timestamp is recorded only after the fetch, so the next device collection cycle retries immediately, flooding logs and spending API budget.

Acceptance Criteria

• #1 A failed MX performance request is attempted at most once per solved mx_performance interval per serial
• #2 A successful request still emits the performance score, and an eligible serial is retried after the interval
• #3 Targeted regression test and repository gate pass

Definition of Done

• #1 just check (ruff format –check, ruff check, mypy, generated-doc drift, offline API conformance, and the marker-filtered pytest run with the 80% coverage floor — this is exactly what the CI test job runs)
• #2 just gen, when metrics, config, endpoints, collectors, the settings schema or the chart config changed — just check includes the drift gate and CI fails the build on it
• #3 Grafana queries in grafana/dashboards/*.json and grafana/alerts/ updated, if a metric or label name changed

Implementation Plan

1. Add a failing per-serial API-error throttle regression test. 2. Record an attempted collection before the API call. 3. Run targeted and repository checks, review, then release.

Implementation Notes

Regression test failed before the fix (second API call after 3s), then passed after recording the attempt before the SDK call. All 45 MX tests passed; just gen changed no artifacts; just check passed 2976 tests with 91.50% coverage. CodeRabbit review pending rate-limit reset.

CodeRabbit completed with zero findings across the three changed code and instruction files. No metric or label names changed, so the Grafana-query condition was not triggered.

Final Summary

Moved the MX performance per-serial timestamp before the API request, so an unsupported or failed request waits for the solved group interval before retrying. The regression failed before the fix and passed after it; 45 MX tests and just check (2976 tests, 91.50% coverage) passed. just gen produced no artifact diff; CodeRabbit reported zero findings.