Unsteady heat support

[marc-flex]

Adding support for unsteady heat simulations.

[dbochkov-flexcompute]

a couple of big picture comments:

• Not sure about having output frequency as a parameter in UnsteadySpec. I think it could be more useful if sampling frequency can be defined for each monitor separately, similar to fdtd time monitors. Say, if I want to have just few snapshots in time for a big monitor, but then also record temperature for each step at a given point.

• For initial condition, I imagine at some point we might need to extend it to non-uniform distribution. While I would say this is out of scope for this PR, I think it would be worth ironing out how that would look like. I see two options

  • initial_temperature: Union[pd.PositiveFloat, SpatialDataArray, UnstructuredGridDataset]
  • initial_temperature: Union[UniformDistibution, CustomDistribution]

  probably the first option is a way to go, unless I am missing something that would require introduction of new specs?

[marc-flex]

Thanks @daniil for looking into this. As per usual these are great suggestions.

Not sure about having output frequency as a parameter in UnsteadySpec. I think it could be more useful if sampling frequency can be defined for each monitor separately, similar to fdtd time monitors. Say, if I want to have just few snapshots in time for a big monitor, but then also record temperature for each step at a given point.

I'll implement this as done in fdtd. I guess my only comment is that depending on the simulation the amount of data generated during the run could be quite big since we'd be saving every time-step and then post-process only the required steps. On the other hand, once the simulation is done and post-processed, all those temp files are removed, so it is only a temporary problem?

For initial condition, I imagine at some point we might need to extend it to non-uniform distribution. While I would say this is out of scope for this PR, I think it would be worth ironing out how that would look like. I see two options...

I like this. This may be related to this issue where we want to interpolate data to a mesh. I guess once I resolve that issue we can revisit this?

[momchil-flex]

Looks great!

The main thing that I think remains (and can't entirely be done now) is imposing some restrictions similar to what we have for fdtd sims. Maybe the only one we can really do now is just setting a limit on the maximum number of time steps in the simulation (and maybe some ball park limits on the time step). Basically users will inevitably create something that has way too many time steps for some reason, unless we limit it.