waddrmgr: fix deadlock

[aakselrod]

This PR fixes a deadlock caused by methods of ScopedManager read-locking the root manager for WatchOnly() and Locked() calls while the ScopedManager itself is locked. This is the reverse order from the normal way of taking locks, but has been hidden by the fact that DB transactions have never happened in parallel until the upcoming postgres changes in LND.

See discussion here for more details.

The first commit updates a test to cause this failure. The following commits, pushed after CI completes/tests fail, contain the fix.

[guggero]

Nice find, LGTM 🎉

[yyforyongyu]

LGTM👍

[losh11]

We found some other places where locking was broken, see:
ltcsuite@69d6e9a
ltcsuite@6570a36

[guggero]

@aakselrod would you have time to create a PR from the two commits linked above?

[guggero]

Thanks for the hint, @losh11. Though I took a quick look at the two commits and it looks like the way @aakselrod fixed the code in this PR (turning watchingOnly and locked of the root manager into atomic.Bool values instead of locking the whole root manager), that should eliminate a whole class of potential deadlocks.
Didn't have time to look into things more deeply though.

I merged the two commits and rebased/fixed them for btcwallet here: https://github.com/guggero/btcwallet/tree/more-locking (in case anyone wants to take a closer look).