NMR: performance issues on demo server

[ml-evs]

I'm seeing some perf. issues on the demo server with the unreleased changes in 0.6.3, seemingly related to passing the completed insitu block through the pydantic model, even with the very restricted example data, it seems to crash the demo server when step = 1... will investigate a bit myself

[ml-evs]

This is not an in situ problem, though it was uncovered here -- out-of-memory due to reserializing a whole bokeh plot multiple times back-and-forth from JSON.