[RTM] FIX: Add second reorient step, in case LIA image is created

[effigies]

This PR adds a second ConformSeries step post-merge, which ensures that the final template is RAS oriented. This reduces to a nibabel.as_closest_canonical, and will not make a new copy (even in memory) if the input is already RAS.

Fixes #636

[effigies]

@chrisfilo WDYT? Now that this stage is three nodes (conform -> template -> reorient), it may also make sense to abstract into a sub-workflow, just to keep the main workflow clean.

Especially if we produce affine transforms to relate the input images to the outputs (#628), which after talking to Satra does seem like a good idea to me, I think it'll be necessary, but that's also a refactoring that can go in that PR.

[chrisgorgo]

I think extracting this is a good idea, but we should leave it until this patter will be used more than once.

As for transformation matrices I asked Satra if he really has a use for them.

[effigies]

Sounds good. Merging.