Fix volumetric MIP

[larsoner]

For volumetric viz and two-sided data in MIP mode, we currently create two volumes: one MIP and one MinIP. This is problematic because of how multiple MIP/MinIP volumes are rendered -- there is no depth sorting / resolution, so which is shown depends on actor addition order.

Turns out there is a way around this: create a single volume and use the A component of RGBA to map to absolute intensity (and map the actual desired voxel alpha elsewhere), as VTK evaluates the A channel to decide "what is maximal" in the volume. There is in principle a slightly better way to do this (that wouldn't require re-baking values) -- "independent component" mode -- but it's broken for MIP. The other possible solution, vtkMultiVolume is broken for MIP.

Comparison, main (blue always wins) by simplifying the LCMV example:

PR (highest voxel along view ray wins):

Also fixes a minor bug with blending="composite", interpolation="nearest" where setting ShadeOn would give a grid outline artifact:

Now fixed:

Changes drafted with Opus 5 and reviewed and iterated on by me.

[larsoner]

pip-pre failures are just pyqtgraph/pyqtgraph#3585 so I'll merge manually (will pin tomorrow if they don't merge my PR ASAP)