Costing improvements

[nilsnolde]

In #3896 I suggested another way of weighting costing factors. We discussed it in more detail last week and found:

1. Re-ordering the factors could make sense
2. Trying to only penalize instead of doing both penalizing & favoring with use_* factors
3. There was smth else but I forgot.. @kevinkreiser @dnesbitt61?

In any case, 1. is not much work by itself, while 2. is quite intense. Both would need to go through a LOT of RAD testing and since they're pretty independent things but very likely causing big changes in responses, we should do one-by-one and separate RAD testing.

It was also concluded that, while strictly speaking we wouldn't have to bump the major version for any of these, it's probably a good idea to phase it in over a period of time, e.g. by providing separate copy/paste/edit costing models or a new next_gen boolean flag in the request, for users to turn on & trial the new logic.