total_rows is not consistent between worker threads and main thread

[nolanlawson]

There's an in-memory cache for total_rows (aka doc_count) in the IDB adapter. This is cute and has some perf benefits, but it also introduces inconsistencies between the worker thread and main thread because we're caching stuff in-memory, outside of the IDB transaction.

Given the increasing usage of web workers and service workers, I don't think this is a safe tradeoff to make anymore. We should just remove this optimization and look for other, safer optimizations.

[nolanlawson]

Technically you can also reproduce this by just using the same PouchDB in multiple tabs, so it's not safe for that case either.