fix bugs, rationalize code, support unix sockets

[thwarted]

• support unix(/path) for go-style DATABASE_DSN
• vim is both more ubiquitous than nano and well known/expected to be the default editor
• remove prev_id in migration plan output, it didn't seem to contribute anything and just made the output confusing
• make sure the saved snapshot sql file ends in at least one newline, as is the POSIX "text file" way
• put a blank line between each table definition in the saved snapshot
• downgrade crashed because it was looking for the no-dump option which wasn't defined for downgrade
• don't try to "figure out" if the operator specified upgrade or downgrade, and then get it entirely wrong, be explicit and indicate that to the planner
• since we're not figuring out if this is an upgrade or downgrade, you can now downgrade to a specific version instead of it trying to upgrade to latest when trying to downgrade (or something, I'm not sure what the hell was happening with this code path, but it didn't do anything sane)
• refactor MigrationRunner plan into subfunctions to reduce cyclomatic complexity by not branching for upgrade or downgrade at every expression and statement
• use u32 instead of usize for migration ids: these are not pointers; 4B migrations is more than sufficient; the platform's pointer size shouldn't influence how many migrations this tool supports
• bump version 0.5.0

[simlay]

If I read this PR correctly, I think you're just supporting tcp and unix as protocols. It may be worth making this an enum. This would make the line if self.protocol == "unix" be more like match self.protocol { Protocol::Unix => {...}, Protocol::TCP => { ... } } which is cleaner. Looking at the list of mysql protocols, unix and tcp are the ones you'll care about so it's probably not worth it.

[thwarted]

I considered that, but decided not to for that same reason.