Technical requirements for modelling centers

## The problem
We need to make some assumptions about what environments we provide tier 1 support for the REF. There are a wide variety of different environments and constraints and we have to what we need to support. This is a catch-all issue to discuss any potential implications for the uptake of the REF within modelling centres.

The REF consists of a large number of small execution jobs with each "execution" within the REF requiring <60s of compute.
The non-HPC based workflow uses celery workers to run tasks which have good support for small jobs, but this may not map well to HPC workflows.
We will produce many small output files and make use of sqlite database for tracking metric executions which may make the REF difficult to use on a lustre FS.

The recommendation would be to deploy the REF via a VM or on a cluster instead of a traditional HPC where possible as this is closer to our intended production workflow, but this may not always be possible.

## Proposed minimum requirements
* The ability to create and use conda environments (python 3.11+)
* linux-64 CPU architecture
* Support FS that is reasonably perfomant with small outputs (JSON, figures etc) (not a hard requirement)
* Access to the datasets from where jobs are being executed
* CMORized datasets
* (Optionally) serving of HTML assets publicly
* We will likely require some inputs from modelling centers for the system-specific configuration needed

We are evaluating some different mechanisms for running these jobs. Options include:
* Running a celery worker via a job
* Producing a Cylc workflow that can then be run

Ideally, we would be able to execute our tasks directly on a node instead of needing to run via a workflow manager, but we will aim to support a Slurm or PBS based workflow (any suggestions for packages to abstract over workflow managers would be useful).

## Additional context

If modelling centers have specific requirements, we welcome contributions to ensure that your users can access the REF

We are capturing additional information about the potential science workflows in #137.

CC @bouweandela