Spike: CLI-side telemetry

Examples of deeper insight we'll gain with client-side telemetry:
- approximate token costs per command
- % of `--jq` usage
- % of JSON output used
- % of usage errors

The client-side telemetry needs to be opt-in. We should:
- design the user-facing part of the experience, and update the design docs.
- decide on the transport/infra, and come up with an epic.

Start by reviewing the existing design docs and the CLI's telemetry-related entry points, if any. Define the opt-in user experience, document it, and evaluate transport and infrastructure options. The work is done when the design docs are updated and a follow-up epic captures the agreed implementation.

Written by the indexing model from the issue text.