GitLab

Made redistribution kernel use 64-bit counts/offsets and implemented a custom all-to-allv to do this

 - diagonal contractions with not all edge lengths equal
 - read_all had a short circuit
 - CTF_VERSION 100 now defined

Also fixed a bug with A*A, which I introduced ~5 commits ago when fixing multi-tensor symmetric contractions.

implemented host buffer allocation into ctr_2d_general and into nonsq_pgemm_test

Fixed and added a test for contractions like C["ij"]=A["ik"]*A["jk"], where the product should not be symmetrized explicitly, since it
is already symmetric due to both operands being identical.

Fixed GNU -O3 error, default parameters were defined in two places for a couple of functions, which is a no-no.

Implemented a scheme that tries to find contraction mappings that keep a tensor frozen to its original layout

Apparently I fudged the mapping algorithm for sum sometime recently while debugging. Reverted to proper mapping search.

upload/download now no longer necessary for scala_backednd. Getting decent single-GPU perf for the benchmark now.