Refactor how process maps are represented and used

[alexdewar]

This will allow new features (e.g. #845), but is also necessary to fix some bugs (e.g. #846).

I think ultimately we should store the individual entries in the CSV files in the Process struct, deduplicating by using RangeMaps (see #874). We will need to transform this data into something usable when we construct an Asset. We should ensure entries are provided for every milestone year upfront, but for other years we should do the check lazily to allow for assets whose commission year is on a non-milestone year (#859).