[dupontbertrand]

This pull request was written and opened by Claude Code, after review by the author.

Adds instrumentation-dashboard, an admin dashboard built on the instrumentation package that ships with Meteor 3.6 (meteor/meteor#14483). A small task app produces real traffic, and /admin shows method calls, p95 latency, error rate and DDP connections, a per-method table, a 60-minute chart and the latest events.

What it demonstrates:

• Listeners only push onto a bounded in-memory queue; all MongoDB writes happen in a periodic flush, so observed methods do no extra I/O.
• Bounded storage: a capped collection for events, per-minute rollups with a TTL index, one heartbeat document per server process.
• Several instances: a compare-and-set on the snapshot's computedAt lets one process recompute the summary each round, and any process can serve the dashboard.
• The dashboard subscribes to one summary document, never to the raw events.
• Private by default: no client IP, and no method arguments except the length of a new task's text.

Verified locally on Meteor 3.6-rc.0: npm run test:headless 8 passing (unit and server integration), npm run e2e:headless 2 passed, npm run lint 0 errors.

Meteor 3.6 is not released yet, so the example targets 3.6-rc.0 and pins ddp-server@=3.4.1-rc360.0: without the pin, rc.0 resolves the 3.5.3 ddp-server, which emits no instrumentation events (meteor/meteor#14849). I will bump the release and remove the pin with the next release.

The root README and examples.json are updated with npm run update:examples.