mempool: remove block even if mempool is empty

[pinheadmz]

Closes #915

Thanks @mafintosh for finding this bug!

Test covers removeBlock with mempool empty and non-empty. Test will fail without code change.

[pinheadmz]

Looking at this some more, I think the bug has actually helped us pass tests that might actually be broken and here's why: As blocks get disconnected, the entire mempool needs to be checked for CSV transactions and premature coinbase spends that are invalid. This is especially likely when disconnecting multiple blocks in a row (a TX may pass insertTX(), but should still be evicted when the next block is disconnected). This logic is written in _handleReorg() but that function is NOT CALLED when a block is invalidated, or if a block is removed for any reason besides a 'reorganize' event, emitted by chain.

I'm going to work on his more before requesting a final review.

[pinheadmz]

So I think there are two more steps that would make this better re: my last comment.

1. Invalidating a block should actually be considered a reorg, disconnecting blocks one at a time and finally emitting a 'reorganize' event. @braydonf we should probably check that the chain-expansion branch does something like this, as opposed to the current behavior for invalidate():

pinheadmz@c827fd0

2. Now that the 'reorganize' event is being emitted, mempool will evict TXs that are now invalid to due a lower chain height (i.e. CSV and coinbase maturity). _handleReorg() already does this, but in a brutal way -- just evicting everything with a CSV or coinbase spend. We can actually do the contextual checks on these TXs and allow them to remain in the mempool if they are still valid:

pinheadmz@cbeb00d

This commit is not complete, TXs that have CSV ancestors will have to be checked as well. See this test in Bitcoin Core: https://github.com/pinheadmz/bitcoin/blob/89a609f182a0f2ff558ed0015753429df9694a18/test/functional/feature_bip68_sequence.py#L302-L313