Inference

This is the current status of the inference code. The objective of this MR is only for review the changes and not to do the actual merge.

There is a remain question about the implementation. The current implementation of the windowed read not read square zones directly, instead read the complete width of the image and a certain number of pixels in height. This is after converted to square patches and then passed to the Network.

Finally the get_lantent method in the MMDC Full Neural Network looks like is not up to date.

PS: I think the entry point for see the logic in the implementation is the follow test: https://src.koda.cnrs.fr/mmdc/mmdc-singledate/-/merge_requests/new/diffs?merge_request%5Bsource_project_id%5D=90&merge_request%5Bsource_branch%5D=inference&merge_request%5Btarget_project_id%5D=90&merge_request%5Btarget_branch%5D=master#fe15da1e28d2857f9d1c879b749a3471ee667a17_0_125

Regards, Juan