Histogram output redesign

[leios]

I want to ...

• [ ] Add an @stamp function that works similarly to an @synchronize in KernelAbstractions. This way, the output does not happen at a fixed stage in the pipeline.
• [ ] Add stamp_by_default argument to kernel kwargs? Maybe better to just always stamp at the end and not give users the option to stop the final stamp
• [ ] Allow for a splat instead of stamp, which will splat out a gaussian with transparency to surrounding areas. This way we can do anti-aliasing without a post-processed gaussian filter. Note that this will force atomic operations and might not be ideal for all cases.
• [ ] Allow for square splats instead of gaussian? No atomic operations needed and this could minimize the number of points for some computation. AD is probably better, though.