Muon

Fixes

> [!IMPORTANT]
> In this release, there are some functions where the bugfixes and improvements led to changes in their behavior. This is why in some instances, this version might produce different outputs when compared to the previous versions. See e.g. 147.

- Enable compatibility of in-place filtering with the latest anndata (148)
- Fix `tfidf` when using data from a layer (147)
- Fix custom chromosome names in `count_fragments_features` (140)
- Prepare to count unique fragments in `count_fragments_features` from the next version (see 110)
- Improve `mu.pl.scatter()`

🐛 → 🦋

**Fixes and enhancements**

* :sparkles: extend _l2norm for sparse inputs (https://github.com/scverse/muon/pull/114)
* improve data type handling for MOFA (https://github.com/scverse/muon/pull/112)
* improve compatibility with the recent scanpy versions (https://github.com/scverse/muon/pull/136)

**Fixes**

- fix handling and saving colour palettes in MuData for categorical and continuous variables (109)
- fix when using sparse matrices in the MOFA interface to combine modalities with missing samples (106)
- fix error messages (95) and mixing metadata and features (96) when plotting across modalities

**Improvements and fixes**

- `mu.pp.intersect_obs` now works for modalities that have no `.X` (93)
- `mu.pl.embedding` now saves the colour palette in `.uns`
- `ac.pl.fragment_histogram` and `mu.pl.histogram` now have save/show arguments (101)
- `ac.tl.count_fragments_features` now has a `stranded` argument
- `ac.tl.nucleosome_signal` now works on more `pysam` setups (105)
- support for numpy 1.24 and newer scanpy versions

**Tools**
- MOFA can be now run in the stochastic mode (SVI) using the new arguments for `muon.tl.mofa` (47)
- Model weights can be visualised with `mu.pl.mofa_loadings`

**Plotting**
- Plotting module has gained new plots such as `mu.pl.scatter` (48)
- It is now possible to define layers as `{modality: layer}` in `mu.pl.scatter`

**ATAC module**
- `pysam` not being installed is now handled more gracefully (46)
- Improvements to the TF-IDF normalisation interface including view handling (88)

**General**
- `sklearn` vs `scikit-learn` dependency issue is fixed (91)

This is a point release with fixes. Consecutive in-place filtering now works as expected (43).