[SMALL] Add support for Undecimated Wavelet Transform, with right adjoint operator

[chaithyagr]

This is a minor PR that will hold the undecimated Wavelet transform class for 2D, which pulls in the transform filters from Sparse2D with help of Modopt.

This is a temporary solution while we wait for general solution from CEA-COSMIC/pysap#106

This PR is particularly important to get the first merge for #35 with cartesian and non-cartesian reconstructions.

[zaccharieramzi]

I think the op and adj_op code need to be refactored. A lot of code is used in common between the multichannel and single channel cases.
For example a op_single_channel and a adj_op_single_channel could be implemented.

[chaithyagr]

Adding, I see that Travis Build also has failed due to missing Sparse2D installations. I will try to get this fixed here as we may need this in the future also.
Note that going with ModOpt based way to obtain MR Filters, we will face this issue, the installations and environment variables must be rightly set.

[zaccharieramzi]

Yes that will have to be explained in the docs somewhere too.
The fix is something like the following:
export PATH=$PATH:/path-to-pysap/build/temp.linux-x86_64-3.6/extern/bin/

[chaithyagr]

Adding @AGrigis to review the travis code updates.

@AGrigis :
Note that this line is temporary:
- pip install git+https://github.com/chaithyagr/ModOpt.git

We are awaiting CEA-COSMIC/ModOpt#64 to be merged to mainline so that we can directly point to CEA_COSMIC master branch.

[zaccharieramzi]

Some questions especially regarding the parallel execution of the wavelet op.

[zaccharieramzi]

LGTM, maybe let's wait for @AGrigis 's input for the travis part, but I am not too worried as this is a hot-fix

[chaithyagr]

Yes, I'll let @LElgueddari just give a quick look at it. Ill mostly talk off-hand with @AGrigis about this.
More importantly right now to get this in, we would need CEA-COSMIC/ModOpt#64 to be in so that I could point to master branch than my github repository

[AGrigis]

I suggest to test against the master banch of pysap and modopt but also from the released version available on pypi.
It can avoid issues with the end user that will probably mix everything.

[chaithyagr]

I suggest to test against the master banch of pysap and modopt but also from the released version available on pypi.
It can avoid issues with the end user that will probably mix everything.

I agree on this, we could use matrix to test with both cases. However, we would need pysap to export the sparse2D binaries to a location for this, (perhaps in /.local/pysap/bin , so that we could do a pip install and point the PATH variable.
I will open an issue in pysap-mri to update travis, and one in PySAP to ensure we export the binaries.

[chaithyagr]

All codes are in. Travis updated to point to CEA-COSMIC master branch of Modopt.
Last checks by @LElgueddari and we can get this to mainline..

[chaithyagr]

Had an offline chat with @LElgueddari , we plan to merge this as there isnt any specific feedback we may require from Loubna for this merge.

Merging to Master