Micom

`plot_fit` got renamed to `plot_association` and will not use LASSO coefficients anymore
as a proxy for univariate associations. Instead univariate non-parametric tests are
run for each metabolite. This should be more stable and avoid issues with low reproducibility
of coefficients. LASSO model performance is still reported as a global measure
of association. In case you need access to the previous `plot_fit` function you can simply
install a prior version of micom in a python or conda environment.

Visualizations are updates requiring less dependencies now and using a current version
of vega-lite.

The transition to the hybrid solver is now completed and the installation does not require
manual steps anymore.

Lowered the iteration limit for the OSQP step in the hybrid solver as more iterations
did usually not lead to an improvement or convergence.

Convert some warnings to debug messages.

Switch the the start method for multiprocessing to "spawn" which should improve
stability for workflows.

Fix a typo in the hybrid settings.

Better default settings for the hybrid solver.

Fixes an issue with taxonomy tables with species rank and without if column.

Fixes yet another annotation bug with AGORA2 due to duplicated exchanges.

Fixes the rank prefix trimming when setting it to False in Qiime2 imports.

Replaces the default OSQP solver with a new hybrid solver using OSQP for QPs and HIGHS
for everything else which improves the accuracy by several orders of magnitude. Note the new
[installation instructions for solvers](https://micom-dev.github.io/micom/installing.html).

The presence or absence of rank prefixes such as `s__` or `g__` for genus and species names,
respectively, can now be detected and resolved automatically by MICOM. This improves
compatibility with Qiime2 and GTDB.

New solver is now tested automatically on all platforms.

Small bugfixes and updates.