Add GRIB save GDT12 (Transverse Mercator)

[DPeterK]

This adds functionality to Iris that allows cubes on a Transverse Mercator grid to be saved to a GRIB2 message. This functionality is contained in a new _save_rules function called grid_definition_template_12.

Please note however that this does not currently work due to a bug in GRIBAPI that means the only key in GDT12 that is specified as being a float32 by the GRIB spec is set as a signed int by the GRIBAPI template. On account of this, _save_rules raises a TranslationError when trying to save a cube with a Transverse Mercator grid to GRIB.

The other changes I've made to get this "working" are:

• This would appear to be the first grid we've saved to GRIB that is not on a simple lat/lon grid. As such I've improved the section of fileformats/grib/__init__ that locates the horizontal dim coords of the cubes to find any horizontal dim coords rather than only Latitude and Longitude.
• The logic for locating the ellipsoid that defines the shape of the earth for the saved GRIB message has been improved. This was done because the previous logic did not cater for a non-Rotated Pole coord system with an ellipsoid defined.
• I've also removed two function calls from latlon_common as it turned out they weren't common to GDT12. Specifically, GDT12 does not define lat/lon for first and last grid points or dx/dy.
• As this PR introduces using only gribapi.grib_set for writing GRIB message key values, I've added a trap for calling this function to the grib save rules tests __init__. The test module itself for this change is very short as calling grid_definition_template_12 will currently always result in a TranslationError.

[DPeterK]

Aah. Removing these wasn't strictly accurate - it's not that GDT12 doesn't set these, it's more that GDT12 calls these keys something different to what GDT0/1 do. I'll push a fix for this presently...

[rhattersley]

The test module itself for this change is very short as calling grid_definition_template_12 will currently always result in a TranslationError.

There's no need for the test coverage to be limited by the bug in the ECMWF GRIB API. Just as you've patched grib_set and friends you can patch grib_get to behave as if it were working correctly.

[rhattersley]

this does not currently work due to a bug in GRIBAPI

Have you raised this on the ECMWF's bug tracker?

[DPeterK]

no need for the test coverage to be limited

Good point, I'll add a commit soon that addresses this.

Have you raised this on the ECMWF's bug tracker?

Not yet but I will.

[rhattersley]

bug in GRIBAPI

I have a work-around on its way...

[rhattersley]

bug in GRIBAPI

I have a work-around on its way...

DPeterK#6

[DPeterK]

This commit fixes an issue with the horizontal grid not being set and adds some tests for saving GDT12.

While I'm waiting for the dust to settle around casting floats as signed ints and casting signed ints as unsigned ints, I've replaced the exception raised when setting the key scaleFactorAtReferencePoint with a warning so that the function grid_definition_template_12 will run.

I've also adjusted the function dx_dy on account of the fact that GDT12 does not have the same key names for dx and dy as GDT0&1. For a similar reason I've adjusted where the function latlon_first_last is called so that it is template-specific.

[DPeterK]

This latest commit has rebased the branch on to all the other changes that have been made since I raised this PR, it makes use of the helper functions from #1480 and #1481 and improves the tests to check that the fixes these helper functions provide are being applied correctly.

[pp-mo]

Since you generalised it, I think the names here are now rather misleading -- they are now really "X and Y" rather than "latitude and longitude".
You should rename the variables, and edit the error message.

[DPeterK]

Good point.

[pp-mo]

All good I think

• apart from all those little picky comments I just threw at it !
  Getting very close now, thanks @dkillick .

[DPeterK]

Review actions - hitting @pp-mo's comments above.

[pp-mo]

Sorted !!
Thanks @dkillick

[DPeterK]

🎉