Honour `_FillValue` attribute when loading ancillary files

Nobody has claimed this yet.

### Are there any linked Issues or Pull Requests?

- ticket:4168 - Original LFRic Trac ticket.
- [ancil:2130](https://code.metoffice.gov.uk/trac/ancil/ticket/2130) - ANCIL workaround Trac ticket.

### Brief description

Presently, the `_FillValue` attribute is ignored when LFRic reads in ancillary files (and presumably other input files). This attribute ([attribute conventions](https://docs.unidata.ucar.edu/nug/current/attribute_conventions.html)) is used to identify missing data] (e.g. to mask out land points for sea-only fields).

### Further details of the issue.

Because the `_FillValue` attribute is ignored, this imposes a constraint on any tools creating files to be read into LFRic that they need to use the model RMDI value as a `_FillValue`.

Instead, the LFRic load should recognise `_FillValue` as representing missing data and translate this to RMDI.

There is an additional complexity in that the netCDF specification (linked above) includes "It is not necessary to define your own _FillValue attribute for a variable if the default fill value for the type of the variable is adequate". So the absence of a `_FillValue` attribute does not imply there is no missing data.

The default fill values can be found most easily with python:

```python
In [1]: import netCDF4

In [2]: netCDF4.default_fillvals
Out[2]:
{'S1': '\x00',
'i1': -127,
'u1': 255,
'i2': -32767,
'u2': 65535,
'i4': -2147483647,
'u4': 4294967295,
'i8': -9223372036854775806,
'u8': 18446744073709551614,
'f4': 9.969209968386869e+36,
'f8': 9.969209968386869e+36}
```

Note that none of these defaults match the LFRic RMDI (the largest finite negative value).

No implementation files, tests, or entry points are named. Start by locating the LFRic ancillary-file loader and its netCDF attribute handling, then inspect how missing values become RMDI. Done means explicit and type-default _FillValue values are recognised and translated to RMDI, with coverage for the supported input types.

Written by the indexing model from the issue text.