Recover from out-of-sync channelmonitors

[tnull]

Matt commented in #11:

What happens if we start and there's channelmonitors that are on different sides of a fork? We need to detect that somehow and get them back to a common ancestor before moving forward with normal block sync.

[tnull]

Tagging this 0.5 as it might finally become relevant after adding bitcoind RPC support. Though, addressing this properly will be a headscratcher as we (do and need to) load ChannelMonitors and give them to ChainMonitor in the Builder, but while only being able to actually connect and poll the RPC ChainSource after/upon start.