Port the speedup from #3163 to the new backend.

Enable cudnn for AbstractConv2d

CorrMM optimizations

Allow optimizer_excluding=inplace to keep the move to GPU.

Clean up (outs|inps)_on_gpu to (outs|inps)_is_tensor

Scan - Move costly checks from runtime to compilation

Better detect if we can use -lblas and more reuse of anaconda blas.

Correctly detect if we can reuse -lblas. We need to check with the compilation argument to be used and execute the program. Otherwise we won't detect some type of error

[ENH] Infer more things in Scan.infer_shape()

Use the right stream for cudnn.

 pep8 tensor signal list directory

Continued work on #3018: Stacktrace fix

Simplify comparisons

Continued work on #3018. Fixed stack trace copy over for additional optimizations, fixed errors in previous commit and inclduded addditional tests.

Implemented tests to check stacktrace is copied over for several opts. Implemented stacktrace copy over for more opts.

[ENG] Doc, blas timing, better error, doc to use pydot-ng, function_dump now strip some stuff.