Port fontc' propagate_anchors.rs to .py; apply it as preflight transformation

[anthrotype]

Fixes #368

This PR translates the fontc' propagate_anchors.rs from Rust to Python and applies the transformation to the GSFont in the 'preflight' step before it gets passed on to the UFOBuilder.
The Rust code was based off the original Glyphs.app's Objective-C code (Georg shared a snippet with us privately) and is as such more 'correct' then the old, reverse-engineered implementation.
The previous glyphsLib.builder.anchor_propagation module is now deprecated but is kept around in case users may instantiate the UFOBuilder class directly with propagate_anchors=True.
Also, the existing propagate_anchors parameters of to_designspace, to_ufos and other high-level methods continues to have an equivalent effect but now controls the new propagate anchors logic. Clients such as fontmake do not need to do anything besides upgrading glyphsLib to use this.

[anthrotype]

I skimmed the regression tests failures (too many to actually vet each single one), and they all seem as expected and for the good.
The only remaining known difference between this Python (and Rust) implementation, on the one hand, and the built-in Glyphs.app's implementation of anchor propagation, on the other hand, is that the former only handles anchor propagation for "master" layers (i.e. those present by definition which have the same layerId == associatedMasterId in all glyphs), whereas the latter also handles it for non-master (intermediate, alternate, color, etc.) layers.
This limitation is due to the fact that glyphsLib does not currently implement the componentLayer property, which would allow to dereference components even when the base glyph does not contain a layer with corresponding layerId (existing non-master layers may get matched by their attributes or new layers get interpolated on the fly): #853 (comment)

However, the same limitation already exists for the current (soon-to-be legacy) anchor_propagation.py, as this is similarly performed for each glyph in each master UFO (for glyph in ufo: etc.), that is only on the default layer of such ufo, and not considering additional ufo layers where Glyphs intermediate (brace) layers are stored by glyphsLib.

So I don't consider this limitation blocking. Implementing componentLayer (with all the required on-the-fly interpolation of missing intermediate layers) is complicated and would prefer not to block on that.

[rsheeter]

I won't claim deep understanding of anchor prop but I don't see any obvious problems and the tests seem to agree, lgtm