Opt-in support for multisampled color rendering

[mosra]

Motivation and Context

As requested by @eundersander on Slack. By enabling simConfig.multisample (exposed to both C++ and Python) all sensors that are just color-only will produce a 8x multisampled output to avoid aliasing.

How Has This Been Tested

There's a new test verifying the output with this flag enabled.

[mosra]

Note: the failing test is due to the lighting setup being somehow different in my working copy, resulting in an image that's uniformly brighter in the parts affected by lighting. I suspect it's caused by something that isn't tracked by Git (some config file different? some scene/stage/whatever file missing, and defaults being picked instead?), and I just couldn't figure out what it was.

To fix that, run the failing test locally with -XS$(pwd) and replace the file it generated.

[jturner65]

Note: the failing test is due to the lighting setup being somehow different in my working copy, resulting in an image that's uniformly brighter in the parts affected by lighting. I suspect it's caused by something that isn't tracked by Git (some config file different? some scene/stage/whatever file missing, and defaults being picked instead?), and I just couldn't figure out what it was.

To fix that, run the failing test locally with -XS$(pwd) and replace the file it generated.

I'll investigate this when I can. If need be I can regenerate the test image.

[aclegg3]

@mosra how big is the ask to make this feature compatible with HBAO? I think the primary value add here is improved optics use cases. That typically means HBAO would be desired as well.

[mosra]

I can do that, yeah -- was basically just waiting for feedback whether you want such feature or not.

I looked into making it work with HBAO initially, but then scrapped that because I wasn't sure if that still counts as a quick MVP that Eric was after :)