LFRic: perform field-copy to full depth of clean halo

[arporter]

Currently any builtin will, by default, iterate to the last owned (or annexed) dof. Although this can be extended into the halo using LFRicRedundantComputationTrans, the depth of the redundant computation is specified at compile time. This does not support the use case of an iterative loop where, to amalgamate halo exchanges, a single, deep halo exchange is performed prior to the loop. At each trip round the loop, the depth of the clean halo is reduced by one. Thus, the depth to which to do the redundant computation is determined at run-time.

Currently, LFRicRedundantComputationTrans accepts a depth option that specifies the literal depth to which to perform redundant computation. If depth is not set then redundant computation is performed to the full halo depth. What we could do with is a third option that says "do redundant computation but only out to the depth of the currently-clean halo."

[arporter]

An alternative approach is to have all built-ins default to iterating out to the currently-clean halo.

[arporter]

Either of these approaches require that we be able to query the field proxy for the depth of the clean halo. This information can be computed but isn't readily available at the moment:
https://github.com/MetOffice/lfric_core/blob/3ef2ace5104efe01580c672e1837a584d428f5af/infrastructure/source/field/field_parent_mod.f90#L123-L131
i.e. we can ask whether a certain depth of halo is clean but not what the maximum clean depth is.

[arporter]

Met with Mike and Chris and agreed this as the long(er) term plan. Mike will add the necessary functionality to the LFRic infrastructure. It will also be necessary to check that the clean halo depth is indeed decreasing for every trip around the loop (it should be but it pays to be sure).

[arporter]

Thinking some more about this, there are other complexities. Currently we have:

so if depth is supplied, it must currently be an integer literal. This information is also used to determine what depth to do a halo exchange, e.g.:

    if (mt_lumped_inv_proxy%is_dirty(depth=1)) then
      call mt_lumped_inv_proxy%halo_exchange(depth=1)
    end if

I don't yet know whether the halo-exchange logic can currently cope with a variable for the depth.