feat(skills): bind performance-slo-planner SLIs to telemetry-foundations vocabulary

Nobody has claimed this yet.

## Summary

Follow-up from the #2420 review (F1, @WilliamBerryiii). Bind the SLIs in `performance-slo-planner` to the shared `telemetry-foundations` vocabulary instead of free-form metric/trace names.

## Proposed changes

- SLIs cite the `telemetry-foundations` conventions (latency = histogram in seconds; `..` metric names; bounded cardinality).
- Add an instrument/unit/metric-name column to the Observability hooks output so each SLI carries its telemetry contract.
- Add `**/.copilot-tracking/performance-plans/**` to the `telemetry-overlay` applyTo glob to mandate the vocabulary.
- For AI-backed journeys, name model latency / refusal rate / accuracy-drift.

## Context

Deferred from #2420 to keep that PR focused on the mechanical and scoping fixes; this meaningfully expands scope.

## Acceptance criteria

- `SKILL.md` references the telemetry-foundations vocabulary for SLIs, and the observability hooks table carries instrument/unit/metric-name.
- The `telemetry-overlay` glob includes `performance-plans`.
- `npm run validate:skills` and `npm run lint:frontmatter` pass; plugins regenerated.

Start with the performance-slo-planner SKILL.md, the telemetry-overlay applyTo configuration, and the generated plugin files. Compare the existing telemetry-foundations conventions and the observability hooks table, then run npm run validate:skills and npm run lint:frontmatter. Done means the SLI telemetry contract, performance-plans glob, and regenerated plugins satisfy the listed acceptance criteria.

Written by the indexing model from the issue text.