Delete code, fix things, go faster. On top of !19560 (closed)
Replaces draw-time bcolor table construction (which only took into account VS and FS) with a global bcolor table. This more or less matches what zink+tu end up doing, but with a bit of cleverness to re-use table entries instead of constantly appending new ones. This moves all of the overhead out of draw-time and into CSO creation time.