Machine readable technical logging

One of the current issues with the UM is that to get technical output from
the model, such as time taken to run, it is necessary to scrape standard out.
This is unnecessarily complicated and fragile.

We should look to implement a flexible system which allows all sorts of
technical details to collated and output in a machine readable format.

Many of these things will come from the top level and will be model wide
but I can envisage the need for per MPI process information as well, so
some mechanism for gathering this also needs to be considered.

Even more likely than per process information is time series information.
For instance we might like to record the number of iterations to
convergence in our various solvers.

We should check to make sure there is not already a library to do this.

No files, tests, or entry points are named. Start by checking whether an existing library supports machine-readable technical logging, then inspect the UM's top-level, model-wide, MPI-process, and solver iteration paths; done means a defined flexible approach covers these outputs without scraping stdout.

Written by the indexing model from the issue text.