Interpolator : Remove overzealous asserts

[johnhaddon]

It's perfectly reasonable to want to interpolate "off the end", and in fact we need to in CurvesAlgo::convertPinnedToNonPeriodic(). I intended to take care of this in #1526 but got carried away and merged before I had done so.

[danieldresser-ie]

I wasn't really thinking about needing this when I reviewed #1526, but if that change is the only place where it's needed to relax this constraint, I'm not really sure if this is really an improvement. Do we have anywhere where we actually want to interpolate off the end of a non-linear curve? That's probably what this assert was originally intended to prevent?

In #1526, interpolator( B, A, 2, C ) is less succinct than just saying C = A - 2 * B, and I'm not sure it's really much clearer.

I also don't see any real problem with merging this, just figured I'd mention that it doesn't feel like an obvious improvement to me.

[johnhaddon]

I'm not sure if you're objecting to the change to CubicInterpolator, or to both. I don't need the CubicInterpolator change, so could drop that. But I can't just say C = A - 2 * B because that's not defined for types like Box, Matrix and Quat, which is what the various Interpolator specialisations were taking care of for me.