Initialize a Cluster currently says that cockroach init "performs a one-time initialization", which may be an insufficient level of detail given that:

1. Users are engineers/DBAs who would likely want to know how/why it works a certain way
2. There is a semi-redundant alternative to just use cockroach start with no --join flag, which is very confusing (IMO) and makes it even more important that we be very explicit / informative about why init exists / how it works / what the difference is

To address this we should update Initialize a Cluster with more detail (including design rationale) from the init RFC.