I manually went through every single dbfake function and ensured it has
the correct lock type depending on whether it writes or only reads.
There were a surprising amount of methods that had the wrong lock type
(Lock when only reading, or RLock when writing (!!!)).
This also manually fixes every method that acquires a RLock and then
calls a method that also acquires it's own RLock to use noLock methods
instead. You cannot rely on acquiring a RLock twice in the same
goroutine as RWMutex prioritizes any waiting Lock calls.
I tried writing a ruleguard rule for this but because of limitations in
ruleguard it doesn't seem possible.