Improve how we apply database migrations

[lutter]

In the hosted service, restarting a node can take a long time. That can be almost entirely attributed to the fact that we rebuild the fdw mappings between shards on every start, and sometimes nodes have to wait for other activity to stop before that rebuild can happen. We rebuild the mappings on every start since we don't properly propagate whether the schema in a shard was changed to the other shards that import tables from that shard.

With this PR, the fdw mappings are only rebuilt if there were actually schema changes, i.e., in the vast majority of cases the rebuild can be skipped and startups will be much faster.

Since interrupting graph-node startup at the wrong time (right after migrations ran but before remapping is done) can cause fdw mappings to not be updated, graphman now has a command graphman database remap to kick off remapping manually.

It also now has a command graphman database migrate to apply schema migrations, though graph-node still does that on startup.