Adapted pars splitting to use Slice, SCAM and DE

Here there are a number of changes that allow for a more general definition of the "slow" proposals when running in parameter splitting mode.

• I've added tests to the demo-basic-multitrial.ipynb notebook. AFAIKS I've not broken (or slowed down) the standard behavior. And everything seems to work reasonably well.
• There is 1 change in the interface: the ChainParsSplitting no longer accepts optional join and split functions but rather the slow and fast parameters masks.
• These changes are needed for using the parameter splitting in GlobalFit.

@maudelejeune it would be great if you can have a look and let me know if it is fine for you before I merge.