Write a `cluster nodes` output parser that's less hacky

[beautifulentropy]

Right now I'm using the CSV parser but splitting on space to capture each column. A side-effect of this design is that we have to throw away the shardslots column. This was probably fine for an MVP but we should rewrite it using a custom splitOnFunc that's less hacky.