Anchor propagation and _isAboveMark interaction

[cmyr]

Ran into this when debugging a diff for Dekko:

• The k_ka-deva is a composite of k-deva and ka-deva. Both of these glyphs have a nukta anchor.
• k_ka-deva ends up with nukta_1 and nukta_2.
• the _isBelowMark check only checks for nukta
• and so this ends up going in abvm, instead of blwm.

There might be some more nuance to this that I'm missing, but it seems wrong that this goes in abvm.

The naive fix would be to do anchor.starts_with("nukta"), and maybe it's that simple?