Doc: Clusters from Scratch: improve index

We've been using the sphinx index markup less than ideally.

Unlike publican, the sphinx index markup goes *before* the text that it points
to, and the sphinx index has only one level of subtopics.

Also, the sphinx index markup (".. index::") offers more possibilities,
most importantly:

• Top-level index term pointing here: single: toplevel_term

• Subtopic index term pointing here (top-level term is not indexed here): single: toplevel_term; subtopic_term

• Two terms, each indexed here as a subtopic of the other: pair: term1; term2

There is room for more improvement but this should correct some little issues
and provide a good base to work with.