Mature Surfaces() class and improve speed on aggregations

The Surfaces() class is designed to be a collection of surface objects (currently RegularSurface() objects).

In some cases we need to compute aggregated values of mean, stddev and percentiles for thousands of surfaces. There is functionalty for this now (pure numpy), but it has to be reviewed for speed and memory efficiency. Some issues:

- [ ] What if one of the individual surfaces are inconsistent with the other. Regridding automatically? Ignoring? Raise Exception?
- [ ] Can it be possible to store a pointer (to file?) each indidual object to avoid loading into memory?
- [ ] Develop methods for memory efficient and fast computation of aggradation stats e.g. running mean; in particular percentiles is non-trivial