With a few minor changes, we could improve performance of simple primitives like
drawLine() and drawRect() for the OGL pipeline:

1) OGLRenderer_DrawLine() currently has a special case for rectilinear lines using
GL_QUADS so that they render consistently according to our spec. With some careful
tweaking, we could use GL_LINES just as we do for diagonal lines. This would improve
our batching potential since we will be rendering all lines with GL_LINES, instead
of bouncing back and forth between GL_LINES and GL_QUADS. Also, GL_LINES requires
half the glVertex() calls and also typically renders as fast as, or faster than
quads on most hardware. (This should also help batching with the drawGeneralPath()
changes from 6322584.)

2) OGLRenderer_DrawRect() currently uses GL_QUADS for the same reasons above. We
could apply the techniques above to cut the number of glVertex() calls in half,
improve performance, and increase batching potential.

3) With 6322584, we have a new operation called DRAW_SCANLINE, which uses GL_LINES
to issue one line at a time. With some minor tweaks to BufferedRenderPipe, we could
probably batch those vertices into one DRAW_SCANLINES operation (similar to
FILL_SPANS), and therefore pass all scanlines as a single vertex array, which
should eliminate method call overhead and improve performance.