Repeating small tests many times was order of magnitude slower. The reason is the computation of
Lagrange polynomials was both inefficient and performed every time an RegularInterpLagrange instance was
created. This causes overhead which is not relevant for large problem sizes but makes running many small
tests slow. The fix for now is to cache the creation of the polynomials. Now they are created only onnce
for each interpolation order, but not once per instance.

Change math markup from mkdocs to sphinx syntax

Made test stricter. The previous parameters used a maximum shift too small to increase
the required margin size, thus not testing anything related to its size.

-Sample rate now specified when constructing operator
-Tolerance now given in same units as shifts, instead of index space.

Limit for coordinate shift was given w.r.t absolute time instead index space.
Bug had no impact because shift was below 1 sample in either case (TODO: larger shifts).

Note results can change within the inversion tolerance. The reason is that
each dask chunk is inverted separately, which may lead to fewer iterations on the
faster converging chunks of data.
The boundary conditions during the interpolation within the fixed point iteration are
flat. All points are used in the convergence measure. In contrast, the old scheme excluded
a hardcoded margin.

TODO: the case where interpolation is set to None is currently not implemented.

All unit tests are passing.

Convergence criterion was excluding margins even though those points were actually valid.

Also, exclude invalid margin points from error measure. The iteration stop criterion
differs from the original one, which used hardcoded margins for this.

Added instrument parameter delay_clock_max giving largest shift allowed to occur during
resampling to different timeframes. For now, this is just a float and not compared to or
inferred from the parameters for the clock drifts / noise.
The interpolation method used is the same as for the interspacecraft delays, but employing the
above limits for the time shift (needed for the internal margin sizes)
Note: the inversion routine for computing the shift between time frames was not touched and
still uses the original interpolation

The electronic delays are assumed to be fixed. Therefore, no new parameters are
needed for the limits. A separate interpolator for electro delays was added to
instrument, using the fixed interpolation method also employed in the adaptive
interpolator for the interspacecraft delays. A parameter check was added to prevent
passing non-scalar electronic delays.

Moved those limits out of the interpolation parameter tuple.

The logic instrument.apply_shift is now in fixed_shift_numpy.AdaptiveShiftNumpy