Rapids-singlecell

cunnData
* added `_inplace_subset_var` & `_inplace_subset_obs`
* added `var_vector` & `obs_vector`
* made cunnData more robust
* made slicing more robust
* enforced `cp.float32` dtype

Fixes
* Replaced deprecated `cp.sparse` with `cpx.scipy.sparse`

Tests
* added tests for `cunnData`
* added test for `decoupler_gpu`
* added test for `gr.ligrec`

Bugfixes and Improvements:
* updates and bugfixes for cunnData
* switched to one PCA - function for cudata and adata
* added `random_state` to PCA
* added layer support for batched hvg-flavors `seurat` and `cell_ranger`
* warning for `cell_ranger` `n_top_genes > dispersion_norm.size`
* added settings dict for and minor bugfix for `normalize_pearson_residuals`
* bugfix for `calculate_qc_metrics`
* bugfix hvg-flavors `pearson_residuals`
* made `max_value` clipping optional for `pp.scale`
* added checks to `gr.ligrec`
* forces float32 for `gr.spatial_autocorr` and fix for `n_perms = 0`
* bugfix for `_check_nonnegative_integers`

New Function
* added `.gr.ligrec` from *squidpy*

Updates
* removed `pp.calc_gene_qc`
* switched `pp.calculate_qc_metrics` to `cp.RawKernel`. Includes functionality from `pp.calc_gene_qc`
* switched `pp.highly_variable_genes` with `flavor='pearson_residuals'` to `cp.RawKernel`
* switched `pp.normalize_pearson_residuals` to `cp.RawKernel`
* added checks cundata.X and cunndata.layers for .nnz < 2**31-1 because of Cupy's indptr being locked to 32bit ints.

Other
* added `.pre-commit-config.yaml`
* added coding-style black

Changes:
* added missing dependency
* changes to typing in `squidpy_gpu`

Changes:
* Added support for spatial autocorrelation (`spatial_autocorr`) with `Moran's I` and `Geary's C`.
* changed some import to be more lightweight

Changes:
* improves `__getitem__` function
* fixed a small bug while creating `cunndata`
* changed to `calculate_qc_metrics` with log1p
* updates `calc_gene_qc` to be consistent with `calculate_qc_metrics`
* updates `filter_cells` and `filter_genes`
* updates docs
* updates Notebooks