Break usage down by project and environment

The usage breakdown ships with request type and SDK (#8186). Project and environment were held back because each costs one request per key, and `usage-data` is throttled at 5/min per user, so a multi-project org gets 429s and a list that ranks the rest wrongly.

Once the endpoint can return the split in one response, add both dimensions back:

- project and environment entries in `BREAKDOWN_DIMENSIONS`
- environment depends on a selected project, so the dimension needs to ask for one first
- a scope that fails must not read as zero usage, and the list must not sit loading when one never answers

The last point covers two bugs found in the first implementation, so they are worth designing out rather than rediscovering.

The removed implementation is in the #8343 branch history, taken out in `4f84749`.

Blocked by #8361.