v0.1.5

improve performance of saveBuffer by writing to memfile directly instead of to a buffer first (also special cases saveBuffer taking an object vs taking a Buffer)
add a CT option -d:FullSets=true/false, which decides how built in sets are stored. If it is true, we store the entire memory of the set (which is more performant, but uses possibly much more memory). Note: loading data works independent of how the set was saved, due to an additional byte to indicate the storage method.

v0.1.4

add save/loadBuffer interface to directly store to or load from a path
handle set correctly and ignore proc explictly (otherwise causes infinite recursion)
improve storing of Tensor by including shape and making sure not to serialize empty tensor