Cellpose

* now install omnipose with pip! `pip install omnipose`
* added PR 416 which removed global logging settings, now turn on logging in a notebook with `from cellpose.io import logger_setup; logger,log_file=logger_setup()`, and from the command line with `--verbose`
* added support for >2^16 masks with np.uint32, if there are <2^16 masks then the masks are returned as np.uint16 still
* fixed bug with torch.long on windows

Fixes to the following bugs:
- `--no_npy` inverted settings
- `omni` not passed through to `remove_bad_flow_masks`

Introducing Omnipose, a collaboration between the Stringer, Wiggins, and Mougous labs written by kevinjohncutler. Read more about it in our [preprint](http://biorxiv.org/content/early/2021/11/04/2021.11.03.467199) and on the Omnipose [README](cellpose/omnipose/README.md). Important new features are:
- `cyto2_omni` model for slight improvement over the 'cyto2' Cellpose model
- `bact_omni` model for bacteria phase contrast segmentation (huge improvement over Cellpose models trained on bacteria, which you can demo with the `bact` model)
- `omni` option to use Omnipose mask reconstruction with your Cellpose model to help reduce over-segmentation (off by default)
- `cluster` option to force DBSCAN clustering in Omnipose mask reconstruction. This is off by default and turned on automatically when the average cell diameter is less than `diam_threshold`. Note theat `scikit-learn` is necessary for DBSCAN, and a CLI prompt will ask you to download it when you run `--omni`.

Several saving options have been included as well:
- `in_folders` saves outputs into separate folders named `masks`, `outlines`, etc. (off by default)
- `dir_above` saves output in the directory above the image directory (useful to have `images` next to `masks` etc.) (off by default)
- `save_txt` turns on ImageJ outline saving (now off by default)
- `save_ncolor` uses kevinjohncutler's N-color algorithm to save masks with repeating but non-touching integers (typically 4 or fewer, 5 or 6 when necessary), which allows segmentations of thousands of cells to be presented without as many colors (which can become very hard to distinguish otherwise). Use in combination with a color map to visualize output.

Several bug fixes and pull requests are included in this release as well.

fixes bugs with

* 2D resizing of flows
* training with CUDA in torch
* `__main__.py` relative imports -> absolute imports

Pytorch is now the default deep neural network software for cellpose. Mxnet will still be supported. To install mxnet (CPU), run `pip install mxnet-mkl`. To use mxnet in a notebook, declare `torch=False` when creating a model, e.g. `model = models.Cellpose(torch=False)`. To use mxnet on the command line, add the flag `--mxnet`, e.g. `python -m cellpose --dir ~/images/ --mxnet`. The pytorch implementation is 20% faster than the mxnet implementation when running on the GPU and 20% slower when running on the CPU.

Dynamics are computed using bilinear interpolation by default instead of nearest neighbor interpolation. Set `interp=False` in `model.eval` to turn off. The bilinear interpolation will be slightly slower on the CPU, but it is faster than nearest neighbor if using torch and the GPU is enabled.

* sped up 3D segmentation by reducing padding
* tile_overlap as a parameter
* fixed bug with batch_size in CLI