Ancillary Variables - Fix concatenate, merge

[trexfeathers]

See #3422 this comment and #3483 this comment

Ancillary variable support should not be released without ensuring concatenate and merge handle AV's correctly

---

Acceptance Criteria

All to be shown working with tests

• cubes with matching ancillary variables can merge to a result containing the ancillary
• likewise.., cubes with matching ancillary variables (not along the axis of concatenation) can concatenate to a result containing the ancillary
• a cube with an ancillary variables and one without (a matching one) will not merge;
  • ... and a descriptive error message is provided, similar to other merge failures
  • ... and the special extra message provided by "merge_cube" describes the precise difficulty
• cubes with differing ancillary variables (i.e. same names but some difference) will not merge;
  • ... likewise, message provisions as above
• likewise.., cubes with/without ancillary variables will not concatenate, and produce appropriate messages
• likewise.., cube with differing ancillary variables will not concatenate, and produce appropriate messages
• cubes with similar ancillary variables which span the axis of concatenation, will concatenate to form a new ancillary variable which spans the concatenated cube
• as all above with cell measures

In short, ancillary variables and cell measures will behave the same as (non-scalar) aux coords.

---

(@pp-mo 2019-11-22) Notes on changes I just made to this ...

• added the "experience-needed: High" label because the existing merge+concatenate code is quite complex, so getting it + proposing changes with confidence is hard !
• added some initial thoughts on acceptance criteria : not necessarily complete; I just updated this while working on something else