blockchain: Avoid deadlock with additional locking

[johnsonjh]

Found via race detector

[johnsonjh]

The Travis failure is a false positive which is fixed in PR #127

[cjdelisle]

pleeease can you just make sendNotification async so that

1. You don't need to lock it
2. You don't need to drop other locks to invoke a send

[johnsonjh]

Probably! I really didn't investigate the logic much in this, just made the patch due to the race detector alert. Leave it open for now, I'll see if there is any reason why not to.

[johnsonjh]

@cjdelisle Not 100% sure if we can - see btcsuite#1563 for the kind of problems mainline has and we don't, and the behavior during deep/extended reorg is a bit of complex. Also btcsuite#1492 and btcsuite#1660, etc. GCash guys have similar changes to what I came up with at gcash/bchd#308 - I'd sort of feel better with this in place until I can get the test harness fixed and can run full regtestnet with the race detector on, and also because I know we had some extended reorg on the chain the other day and my node with these patches made it through and some of the gridnodes running older software crashed ... but recovered after restart.

Fixing up the test harness is going to high priority for me, so I can more easily test these edge cases, especially for some of the more invasive changes I have in testing like the UDP FBT, etc.

[johnsonjh]

• Verified clean merge with recursive strategy against develop as of Thu Nov 26 02:01:32 PM EST 2020 + PR Improvement in bbolt/walletdb speed #111 + PR More reliably detect if we are current (or not.) #142 + PR Fix bug where pktctl -h would show help twice. #151

Merge made by the 'recursive' strategy.
 blockchain/accept.go |  2 ++
 blockchain/chain.go  | 17 ++++++++++++-----
 2 files changed, 14 insertions(+), 5 deletions(-)

[johnsonjh]

• NOTE: Applied to WIP rebase branch successfully.