Evalutils

- Pins Pandas version due to a bug with dataframe conversion to
dictionaries in Pandas 0.24.0
- Updates the export.sh template to gzip the docker container after
creation
- Added Python 3.7 support

- Adds `evalutils.stats` for calculating
common metrics in medical imaging
- Adds **experimental** Windows support - we do not have CI on all of
the windows tests so please report any errors
- Adds `evalutils.roc` for calculating
bootstrapped roc curves

- Adds Segmentation and Detection challenges to the project generator
- File loaders now return lists of dictionaries rather than
dictionaries
- Renamed `intersection_over_union` to
`jaccard_index`
- Improved image memory management. Added
`io.ImageLoader` and separate
`load_image` and
`hash_image` functions.
- `score_case` is no longer a static
method

- Fixed a bug where the number of
`false_negatives` could be less than
0.

- **Breaking change:** Renamed `Evaluation` to
`ClassificationEvaluation`
- Adds support for `DetectionEvaluation`

- Simplifies the example template