support for force-master-failover command

[shlomi-noach]

This PR introduces the -c force-master-failover command.

This tells orchestrator to assume the master is dead, and that it should kick a full failover, by whatever means it finds necessary to promote a replica.

Usage:

orchestrator -c force-master-failover -alias mycluster

or

orchestrator -c force-master-failover -i some.instance.in.a.cluster

orchestrator will figure out what the master is, shut its eyes and make pretend it's dead, then kick a failover.

Contrast with force-master-takeover where the user specifies the replica which must get promoted (more control but potentially more complicated for the user).

Note: this PR extends #196

cc @github/database-infrastructure

[shlomi-noach]

Web interface will look like this:

[jonahberquist]

Nice! That's a good interface update, too.

[shlomi-noach]

Web interface simplified. For master, the "recover" options look like:

For intermediate master:

I completely removed the recover-lite options, as well as various "regroup" options. I'm htinking they're not really interesting, and are mostly confusing at time of failover.

[jonahberquist]

Even better 👍