Deprecation of `scipy.interpolate.interp2d`

[mlincett]

From current unit testing output:

DeprecationWarning: `interp2d` is deprecated!
`interp2d` is deprecated in SciPy 1.10 and will be removed in SciPy 1.12.0.
For legacy code, nearly bug-for-bug compatible replacements are
`RectBivariateSpline` on regular grids, and `bisplrep`/`bisplev` for
scattered 2D data.
In new code, for regular grids use `RegularGridInterpolator` instead.
For scattered data, prefer `LinearNDInterpolator` or
`CloughTocher2DInterpolator`.

Here are the occurrences:

flarestack/core/llh.py:        f = scipy.interpolate.interp2d(dec_bins, gamma_bins, acc.T, kind="linear")
flarestack/icecube_utils/reference_sensitivity.py:    sens_ref = interp2d(np.array(sindecs), np.array(gammas), np.log(sens.T))
flarestack/icecube_utils/reference_sensitivity.py:    disc_ref = interp2d(np.array(sindecs), np.array(gammas), np.log(disc.T))
flarestack/icecube_utils/reference_sensitivity.py:    sens_ref = interp2d(np.array(sindecs), np.array(gammas), np.log(sens.T))
flarestack/icecube_utils/reference_sensitivity.py:    sens_ref = interp2d(np.array(sindecs), np.array(gammas), np.log(sens.T))
flarestack/utils/percentile_SoB.py:    spline = scipy.interpolate.interp2d(x, y, np.log(ratio))

[mlincett]

As of SciPy 1.14.0, this now raises an exception. We should pin scipy at 1.13.1 in pyproject.toml, as long as the interpolator is not replaced.

Also, note that there have been and still are some reported performance regressions so if we try to upgrade we may want to benchmark performances.