Fix iris3 coord units

[pp-mo]

WIP: DO NOT MERGE
until we have Iris3, and this is re-tested + OK

Closes #207

These changes will be needed when we release the changes on iris default_units feature-branch
(but which is not yet merged, hence WIP flag).
They just ensure "units=1", for unchanged behaviour when creating coords.
That is anyway benign : it all works with current Iris.

( checked out locally, against the Iris feature-branch current content )

[coveralls]

Coverage remained the same at 92.482% when pulling 0e2afbc on pp-mo:fix_iris3_coord_units into 4b7797f on SciTools:master.

[pp-mo]

Outstanding Questions :

• model_level_number now has "no_units" in Iris PP loading - is that right??
  • - ok, checked that units=1 is expected for both PP and netcdf data in Iris test CDLs iris/tests/results/...
• satellite series/number/type numbers : units=1, maybe some should be "no_unit" instead ?
  • - not so clear, maybe needs checking against satellite knowledge : can we see some netcdf coded data of the same type ?

[pp-mo]

Added 2 new commits to fix for recent changes in Iris : Notably from SciTools/iris#3804, which has rationalised types of coordinate-system properties (most are now cast to floats)

This tested all passed with this recent Iris commit

[pp-mo]

Now done in #231