improve fill() routines

[irmen]

The default buffer size for gfx2.fill() is 64 entries which is far too small to reliably fill all polygons, especially in high res screen modes.

Fix this maybe by using the stack/queue code in #145

Don't forget monogfx.fill() as well.

Another optimization that could be made is to remove the separate calls to horizontal_line, and instead immediately pset() the pixels after the pget()'s. Might even optimize that by using both vera data ports 1 for reading 1 for writing, with appropriate auto increment/decrement.

see discord DM from leroy for code

[irmen]

Note: the monogfx optimization work is done in the following branch: monogfx_fill_optimization

gfx2 has been completed (performance wise, not the limited stack issue)