Minor fixes

Some doc / tutorial updates

Fix in scan_merge: clone each inner graph independently

Add compatibility unpickler for python 3, use it in test

[MRG]Conv3d2d

Faster opt

theano-nose theano/tensor/tests/test_blas.py:t_gemm.test0a

Fix test of shape_of_variables and raise a good error to explain that the FunctionGraph interface changed.

Disable the clone in some tests and other place where it should not be done.

Don't make TopoOptimizer profile the validate time as this is not needed.

This lower memory usage and speed up optimization by asking for less merge.

This make Merge Optimzer take 21s instead of 35s in one case.

I think this work correctly in case of revert.