Refactor LISI metrics (#395)

* improve np.nansum and use njit for HBeta

* use np.eye function for convert_to_one_hot and reduce redundant initialisation

* use with loop for temporary files

* Add importlib-metadata in setup.cfg

* set minimum python version to 3.8

* python version inas strings

* update Scanorama version

* fix n_batch

.github/workflows/test.yml

@@ -73,7 +73,7 @@ jobs:
         runs-on: ${{ matrix.os }}
         strategy:
             matrix:
-                r: [4.2]
+                r: ['4.2']
                 python: ['3.10']
                 os: [ubuntu-latest]
         steps: