EG: SlideGraph Pipeline As Advance Example Usage (#180)

Finally something to bind all the parts in the toolbox together for a single purpose.

- [x] Framework conversion
- [x] Inference integrity check
- [x] Training integrity check
- [x] Add cell compositions
- [x] Explanation  

There are several parts / considerations that may need to be split into / implemented by other PR during the making of this notebook to reduce its complexity

- Generating some specific train/valid/test split
- Drawing generic graph
- Drawing graph by a specific method
- Classifcal feature extraction
  - Cell compositions of patches
  - Nuclei morphology
- Extending some sklearn with `partial_fit` as fitting 1mllion ResNet50 vectors is impossible (OOM)
- Full training engine (may be far into the future)

Lastly, full pipeline as a notebook is bad for organizing code. Perhaps, we should maintain a subfolder called `pipeline` within `examples`, such as `examples/pipeline/slidehgraph` will contain `.py` code of slide graph (parts split/merge from notebook). Additionally, due to the complexity of making a pipeline, tests should also be written, however, notebooks won't facilitate that easily. These tests will focus solely on behavior tests i.e at least being able to replicate the original paper results.

Co-authored-by: Dang Vu <[email protected]>
Co-authored-by: David Epstein <[email protected]>
Co-authored-by: wenqi006 <[email protected]>
Co-authored-by: John Pocock <[email protected]>
Co-authored-by: Shan Raza <[email protected]>

Author: 5 people