Rainfall velocity is a hacked diagnostic

[tommbendall]

The rainfall velocity is a useful diagnostic when running with a rain variable, but getting it from the physics scheme into the field container is a complete hack and needs fixing! This could easily arise again if we have auxiliary fields in physics schemes that we would like to output.

[tommbendall]

I think a good solution to this would be to allow TimeDiscretisation and Physics objects to hold a list of diagnostic fields, which would be default be empty, i.e. in the base class __init__ methods:

self.diagnostic_fields = []

We could then add default diagnostics, such as rainfall velocity or mr_sat to this list. (I wouldn't intend on adding anything by default to the time_discretisation but want to put the capability in there to help with debugging in the future)

We wouldn't want these default diagnostics to be output automatically though -- we want the user to specify them in the normal way as name/objects in the list of diagnostic_fields passed to the IO.

When the time stepper tells the IO to setup_dump, we should then have the IO loop through all physics schemes / time discretisations to check for diagnostic fields. They can then be registered to dumping at the appropriate moment.

---

I think this is a fair bit of work but wanted to get my thoughts recorded.