Generate differs in 10 and 100 runs

New "new" format
Init must check if both files are the same; if so, new format; if not, new new format.
Validation needs pairs of files (Generated/Reference); assume short is always present, long is optional
"update" can read the Validation directly
If the generated file doesn't exist in the long run, ask again
In "init", update the reference (this should be done inside Generator)\

Not reducing

caseSetupDict.static, maybe?
When there is no reduction, there is no need to build the short/long run things
- The old format supports that, we just need to figure out when there are no reductions in init to generate the old format.
- There is also a suggestion to change the run-dir when there is no reduction, which I'm not sure if possible due the other of "expander"/"reducer"...
- Also, would need to have, in meta information, to have both (one option) rundirs.
  - This would help the other point of keeping the run-dir static for init, update and run.
  - Expanding all run-dirs at the start would save a few runs too.
In the pipeline, the sequence is Expander, Reducer, Loader.
- The Expander doesn't know if the case can be reduced or not, and can't make a judgement if we need 1 or 2 run-dirs.
- Maybe Reducer could have a "reducer probability" kinda of thing that Expander could call to find out the number of run-dirs.
  - Function returns: caseSetupDicts that can be reduced but are not continuations, caseSetupDicts that are continuations and caseSetupDicts that can't be reduced. I'd like to make it an iterator, but I guess returning a Vec is good enough (specially since we need to sort the files in the filesystem to find out the continuations).
- Should we follow the same idea for controlDicts?
Need to document this properly