RB for WINTER

[robertdstein]

This PR adds support for realbogus in WINTER. The model itself is defined in a new external package (https://github.com/winter-telescope/winterrb), which is added as a dependency. The model weights are downloaded from the github directly, so can be easily bumped by pointing to a new URL.

This is scalable to any other pytorch model. Fix #877.

The PR also adds cuts to what we sent to IPAC, in particular rb > 0.5 or null. It also adds an sgcut on the drp side. However, these changes only impact what is sent. There is no change to the way candidates are initially generated. After testing RB in deployment for a while, we can consider a loose rb cut before entering things in the database.

[coveralls]

Pull Request Test Coverage Report for Build 9099585681

Details

• 78 of 82 (95.12%) changed or added relevant lines in 7 files are covered.
• No unchanged relevant lines lost coverage.
• Overall coverage increased (+0.1%) to 84.646%

---

Changes Missing Coverage	Covered Lines	Changed/Added Lines	%
mirar/paths.py	3	4	75.0%
mirar/processors/sources/machine_learning/pytorch.py	52	55	94.55%

Totals
Change from base Build 9056061846:	0.1%
Covered Lines:	10949
Relevant Lines:	12935

---

💛 - Coveralls