Speed improvement: pre-interpolate geometries when using `fixed_time_step`

`fixed_time_step` loop calls the geometry provider's time-interpolation once per step. As a result, loading multiple geometry time slices pays the interpolation cost on every step rather than once at the geometry initialisation, inflating both compile time (~2-5x) and run time (scaling with step count).

However, with a fixed dt, every future time point is knowable in advance, so we could in theory make `fixed_time_step` pre-interpolate all of the geometries.

Locate the fixed_time_step loop and the geometry provider's time-interpolation call mentioned in the issue. First measure the current compile and run-time costs for multiple geometry time slices, then trace geometry initialization and fixed-step time points. Done means interpolation is performed once where appropriate while simulation results remain unchanged and the reported costs improve.

Written by the indexing model from the issue text.