Re: Fine-grained caching

[Chris Warburton]

As a first step, we can separate the cache-reading from the result-calculating:

 - The generate functions are the only place where we calculate the results.
 - The generate functions are only called from their scripts (and tests).
 - The normal function checks for a cache file, like now, but aborts if not
   found.

This way, we can force the expensive calculations to only be run once, while
generating the cache, and for everything else to use the cache.