Avoid calling into AnalysisManager within per-pixel loop

[philipbelesky]

Calling into the AnalysisManager to get lookup pixel colors is significantly slower (across the loop's course) than the previous method which had a direct lookup of a color table which was local to SolveInstance(). Ideas for improving performance here while keeping the abstractions from the analysis classes:

• Extracting the relevant lookup tables into a local variable before entering the loop
• Setting the relevant GetPixelColor for each analysis type as a delegate method before entering the loop

I'm not very clear as to the relative low-level performance differences between different options here so will play around and see what works best. As we are dealing with pretty small differences in time I might try and make some mock functions that can simulate the process of coloring the per-pixel loop in a way that can be more easily run multiple times and averaged. I had a go at using benchmark do net for this, but didn't have any luck integrating it as either a console app (which then couldn't call into rhino common) or as a separate plugin (which then couldn't instrument methods within SolveInstance). Maybe the dotTrace profiler will help.