Only expose final results

[Chris Warburton]

Currently, we have a bunch of modules which intermingle utility
functions and data (i.e. memoised thunks).

Split these up, so we have some modules which provide (mostly pure)
utilities, which can be tested with dummy inputs, and others which
provide data, generated using those utilities.

Anything that's only required in order to generate the data, should only
be accessible from within that data-generating module. That way, a
"downstream" module like e.g. conjectures, will only have access to the
actual data, and some helper functions for acting on it, rather than a
bunch of needlessly-exposed data-generating functions.