R26

Chat reset is fast again.
Fixed: Regenerating a message while reloading a model would cause an exception.
Using llama functions directly where possible to reduce dependence on the common library.
More little fixes i cant remember.
(Bugs and software development is like playing whack-a-mole... only a mole pops out of every other hole)

And as usual... Use the CUDA backend plugin from the R17 release.