Implement our own Redshift "classic resize"

[geoffxy]

Redshift's classic resize can be a disaster (it has run for 15 hours+ before without any progress on a 6 GiB cluster). We cannot realistically make any predictions about how long a classic resize will take because it depends on factors we have no control over nor visibility into.

An alternative is to implement our own manual "classic resize". We create a new cluster of the intended size and then copy data into it.