feat: Adjusting volume rendering default number of samples

Extracted from https://github.com/google/neuroglancer/pull/854 as I'm changing the scope of that one.

The default number of samples during volume rendering seems to be consistently too low. There are two changes we could make for this. The first, I think, is not likely to cause much issue and is pretty safe to do. That is to change the number of samples that a layer gets in the new multi-channel image setup route. This is because it doesn't have any impact on existing neuroglancer states, it only takes affect for new states. For this path, currently we take 256 samples. But we could bump to 512, or 1024 for example.

The other path is trickier, which is the default that any neuroglancer image layer which has not modified the number of samples previously would get. If the number of samples is always too low, then we could consider also bumping this. This would have an effect on existing states for neuroglancer that never changed the default volume rendering samples though. Since volume rendering is still listed as experimental, that might be an ok change, but I do understand it is perhaps a bit more controversial a change than the one to the multi-channel only setup. Currently we take 64 samples, which could be 256 for e.g.

Overall, I think the real aim would likely to be to improve this system with mixed LOD rendering during volume rendering, but in the interim, this could be a useful change.

We've also discussed how for the current non-multi resolution volume rendering you actually might want to instead fix a resolution level instead of number of samples.