Add support for context execution affinity parameters (limiting SMs in use) #747
Description
CUDA 13.0 introduced "execution affinity" parameters, which, for now, are a setting of the number of SM's which the context is allowed to use. We need to find a way to add this to our APIs. For now, we should not bother the user with the contrived dynamic array of params of multiple types etc. etc. - it's just an unsigned int value, or a specification that it's the "max" number of SMs - which we will assume is the equivalent of not setting this at all.