Add `aligner` notebook

[arentkievits]

Currently, only a stitcher notebook is available to do the montaging, but we also need to do 3D alignment, hence the task to write the notebook...

• [x] Imports
• [x] Create downsampled stack
• [x] Get downsampled section pairs (basically adjacent sections in z)
• [x] Get point matches
• [x] Filter points matches (exclude point matches at edges of stack)
• [x] Rough alignment (BigFeta)
• [x] Create roughly aligned stack when satisfactory
• [x] #5
• [x] Alignment files
• [x] Get tile pairs in z
• [x] Get point matches in z
• [x] Fine alignment (BigFeta)
• [x] Generate stack and upload to render-ws

[arentkievits]

Problem: pointMatchClient mostly finds point matches at stack bounds, where the intensity drops to zero. Clipping intensities and reducing bit depth reduces this, but not a lot of matches are found. Therefore, point match searches should be restricted to regions within the stack bounds and be more strictly defined. Define a mask Clipped

[arentkievits]

To come to a more robust results, it's probably better to split the point match candidate finding and RANSAC. To do this, use routines in features.py. I've now added an extra notebook align.ipynb for this purpose, align_render-client.ipynb will use the standard PointMatchClient and continue with these results. Filter point matches from PointMatchClient based on proximity to stack edge. Come back to align.ipynb later...

[arentkievits]

Rough alignment works now, but apply ds rough aligned transformation to point matches to visualize overlay of tilepairs with point matches. Give more insight in correctness of rough alignment