Sleap

Alpha release of 1.0.4.

- Hotfix for CLI use of topdown batching (`sleap-track [...] --topdown.batch_size`)

Alpha release of 1.0.4.

- Add optional batching to topdown predictor inference for speed up in certain cases. This can enabled by adding `--topdown.batch_size 32` to the `sleap-track` CLI.
- Modified performance logging to report examples/sec and frames/sec separately (e.g., for topdown).
- Add version number to GUI title.

Minor enhancements:
- explicitly require numpy>=1.16
- check for non-nan points when adding predictions
- remove empty frames when deleting instances
- command to add user instances for all predicted instances

Minor improvements:
- use "auto" as default crop size in baseline topdown configs
- expose early stopping params in GUI

Bug fixes:
- pin attrs to 19.3 (fixes 346)
- pass original image through topdown inference pipeline to flow tracker (fixes 343)

=====

=====
Validation set:
=====
Frames: 1 / Instances: 2
Metrics:
Dist (90%/95%/99%): 1.377530428143121 / 1.4410399538615883 / 1.4918475744363626
OKS VOC (mAP / mAR): 0.8252475247524752 / 0.85