Might be nice to parallelize PNG image loading

[smcameron]

This would require separating the PNG loading/decoding step from the opengl texture binding stuff, since the opengl stuff can only be called from the main thread, and thus is serialized. I believe most of the time loading images is actually spent in PNG decoding, so separating this into a first step in which all the PNG coding is done (in parallel) and a 2nd step where these loaded PNG images are made known to opengl (serially) should in theory result in significantly faster load times.