change arg order of Copy/Memcpy to follow inputs-then-outputs convention

instead of C memcpy order -- from (dst, src, n) to (n, src, dst)

remove failing repeated arg death test after #7

removal of check (PR #7)

Tensor constructor: values arg is const

allow Tensor with empty dims (a scalar)

shapes and other use cases)

use in loss functions for scalar loss output

add system/anaconda dirs to build_env.py, filter non-existent dirs

dirs

(2) blob serialization comments
(3) cudnn: putting it under a separate device name
    so we can explicitly choose cudnn instead of
    having CUDA device prioritizing it.
(4) note that mint is not available with ipython
    due to zeromq conflict
(5) db_throughput utility
(6) added gprofiler

[misc] Update license and readme.

[binary] some enhancement to move over caffe databases.

[python] added an alexnet2 example notebook.