[Wallet] Avoid using re-entrant locks

[tmpolaczyk]

We have already found a few hard to debug issues because of re-entrant locks in the wallet. This happens when one thread calls RwLock::read one time, and then calls it again without dropping the first handle. This operation will succeed as long as there are no writers waiting for the lock (on Linux, this behavior may change in other platforms).

The simplest solution would be to replace the RwLock with a variant that panics on re-entry. This way we would convert the deadlocks into panics. Unfortunately I don't known of any library that provides something like that, so we would need to do a quick assessment.

The other solution would be to re-think the architecture of the wallet so that the RwLock is not needed, this way we would avoid the problem of deadlocks entirely.