call FETensors each time step

[tdcwilliams]

could help with some weird stuff

• eg thickness "blobs" for some parameter choices.
  Would also
• make the node positions and velocities at the nodes exact
• make drifters more accurate
• values of tracers more accurate (M_surface should change too)

TODO
call FETensors() and also calculate M_shape_coeff and M_surface each time step

[einola]

Just to clarify: The physical advantage of doing this should be weighted against the numerical cost. Numerical cost was the reason for doing it the way it's done now ... but it's ok to reconsider that if it's worth it.

[tdcwilliams]

Repeated a bad run with 100m blobs: 12kPa-h^2-tp10

• compression_factor 12000, exponent_compression_factor=3, topaz relaxation time 10days
  FETensors 1.77% of total
  total (21 days): 9h24min

Haven't got the timings for the identical run, but here are some others
6kPa-h^2-tp15
auxiliary 0.01% of total
total (21 days): 9h21min

6kPa-h^3-tp15
auxiliary 0.01% of total
total (21 days): 9h19min

So while the fraction for FETensors has increased from 0.01% to 1.77%, the total time has not increased too much.

The run hasn't completed a full winter yet but there are no signs of any blobs, while the old code had a few already. The 100m blob didn't appear until a bit later so I'll let the new run continue for a bit longer to make sure they're really gone.