Correct real constants for ENDGame

# 🐜 Bug Report
For ENDGame, we may be saving incorrect real constants to F03 ancillaries. But our understanding is that the UM only uses the real constants for (obsolete) New Dynamics grids, so this issue may be harmless. See old trac issue 854 for more detail.

Given the current behaviour has been in place for 8 years (or more), care should be taken that any updates don't break other tools (UM, obviously, but also mule, etc). It may be better to document the current behaviour rather than fixing.

From that ticket:

> I'm afraid I don't think that is correct for ENDGame. The BZX+BDX and BZY+BDY values refer to the first grid point for whichever stagger location the field is on. They will vary between P, U and V grids. In the case of ND this happens to be a P point, but for EG it is offset by half a grid spacing.
>
> For the EG case notice that the grid origin is half a grid spacing south west of the first P point. I think what you need is (0.5 * BDX) + BZX and (0.5 * BDY) + BZY.

Start by locating the ENDGame handling that writes real constants to F03 ancillaries and compare its BZX/BDX and BZY/BDY values across P, U and V grids. Review old trac issue 854 and check the UM and mule compatibility concerns. Done requires a clear decision to correct the constants or document the existing behavior without breaking those tools.

Written by the indexing model from the issue text.