This bug is about adding telemetry probes to track the number of cache load failures we experience. We want to be able to differentiate between:
- SQLite error
- metadata inconsistency
We also probably want a histogram over the number of "accessed" origins that we had to reinitialize the slow way. (So each data-point is a separate call to LoadQuota).
We probably also want telemetry to know at startup which of the following happened:
- We had a valid cache and we tried to use it
- We had a valid cache but the "purge cache" mechanism forced us to not use it.
- We had just upgraded so there couldn't be a cache. (Not sure if we need this, but it helps provide total coverage.)
- We didn't have a valid cache
(This was suggested by :asuth in D38817)