Precalculate bi/tri-cubic coefficient inverse in cubic interpolators

[CnlPepper]

The current cubic interpolators are unnecessarily re-inverting the coefficient matrix for each cell. This can be precomputed. This would also improve the numerical accuracy of (beyond 5-6 decimal place) by avoiding an unnecessary pre/post normalisation that occurs around the inversion.

[jacklovell]

We now recommend the use of Raysect's interpolators, which don't have this issue. All internal uses of the interpolators in Cherab were swapped to Raysect ones in #304. The existing interpolators have been marked as deprecated and retained only for backwards compatibility.

The other place this issue appears is in the caching objects. These are being re-written in #363.