Is it worth making better use of compressed Jacobian format?

Transferred from [ukca:#326](https://code.metoffice.gov.uk/trac/ukca/ticket/326) on MOSRS.

The Jacobian in ASAD is a sparse matrix, which is stored in compressed format as `spfj`. Previous profiling work has indicated that the majority of the computational cost is in the nested loops in `splinslv2` (within `asad_sparse_vars.F90`). However, these loops do not make use of compressed formats when looping over entries of the Jacobian and instead loop over *all* entries of the sparsity pattern and check whether each is zero.

On branch [r7184_lu_refactor](https://github.com/joewallwork/ukca/tree/r7184_lu_refactor), I implemented compressed-sparse-row (CSR) and compressed-sparse-column (CSC) storage formats for the Jacobian and made use of them in the nested loop structure. There doesn't appear to be a noticeable runtime difference in my testing so far (at N48 resolution).

We should investigate whether there's a performance enhancement in higher resolution cases.

Start with the nested loops in splinslv2 within asad_sparse_vars.F90 and review the compressed Jacobian work on branch r7184_lu_refactor. Benchmark the existing and compressed formats at higher-resolution cases, since N48 showed no noticeable runtime difference. Done means establishing whether compressed storage provides a measurable performance enhancement and documenting the result.

Written by the indexing model from the issue text.