Plotting UMAP in backed mode leads to error when setting color palettes

• [x ] I have checked that this issue has not already been reported.
• I have confirmed this bug exists on the latest version of scanpy.
• (optional) I have confirmed this bug exists on the master branch of scanpy.

_set_colors_for_categorical_obs copies adata if it is a view (e.g. rows/cells are reordered for plotting), leading to error in plotting.

The below fails as adata is view (reordered cells) and color of 'CellType' is not in uns
However, if we first run the plot without reordering cells so that uns color is set on adata and then reorder the cells and plot the view, this works

random_indices=np.random.permutation(adatas['Human'].obs_names)
sc.pl.embedding(adatas['Human'][random_indices,:],ACTIONet2D',
                color=['CellType'],hspace=0.9)
---------------------------------------------------------------------------
ValueError                                Traceback (most recent call last)
<ipython-input-48-8a1d75e9b375> in <module>
      2 np.random.seed(0)
      3 random_indices=np.random.permutation(adatas['Human'].obs_names)
----> 4 sc.pl.embedding(adatas['Human'][random_indices,:],'ACTIONet2D',
      5                 color=['Batch','Region','Condition', 'Grade', 
      6                        'PMI','Sex', 'Age','SubType','CellType'],hspace=0.9)

~/miniconda3/envs/csp/lib/python3.8/site-packages/scanpy/plotting/_tools/scatterplots.py in embedding(adata, basis, color, gene_symbols, use_raw, sort_order, edges, edges_width, edges_color, neighbors_key, arrows, arrows_kwds, groups, components, layer, projection, scale_factor, color_map, cmap, palette, na_color, na_in_legend, size, frameon, legend_fontsize, legend_fontweight, legend_loc, legend_fontoutline, vmax, vmin, vcenter, norm, add_outline, outline_width, outline_color, ncols, hspace, wspace, title, show, save, ax, return_fig, **kwargs)
    250             groups=groups,
    251         )
--> 252         color_vector, categorical = _color_vector(
    253             adata,
    254             value_to_plot,

~/miniconda3/envs/csp/lib/python3.8/site-packages/scanpy/plotting/_tools/scatterplots.py in _color_vector(adata, values_key, values, palette, na_color)
   1267         color_map = {
   1268             k: to_hex(v)
-> 1269             for k, v in _get_palette(adata, values_key, palette=palette).items()
   1270         }
   1271         # If color_map does not have unique values, this can be slow as the

~/miniconda3/envs/csp/lib/python3.8/site-packages/scanpy/plotting/_tools/scatterplots.py in _get_palette(adata, values_key, palette)
   1238     ):
   1239         #  set a default palette in case that no colors or few colors are found
-> 1240         _utils._set_default_colors_for_categorical_obs(adata, values_key)
   1241     else:
   1242         _utils._validate_palette(adata, values_key)

~/miniconda3/envs/csp/lib/python3.8/site-packages/scanpy/plotting/_utils.py in _set_default_colors_for_categorical_obs(adata, value_to_plot)
    468             )
    469 
--> 470     _set_colors_for_categorical_obs(adata, value_to_plot, palette[:length])
    471 
    472 

~/miniconda3/envs/csp/lib/python3.8/site-packages/scanpy/plotting/_utils.py in _set_colors_for_categorical_obs(adata, value_to_plot, palette)
    428         colors_list = [to_hex(next(cc)['color']) for x in range(len(categories))]
    429 
--> 430     adata.uns[value_to_plot + '_colors'] = colors_list
    431 
    432 

~/miniconda3/envs/csp/lib/python3.8/site-packages/anndata/compat/_overloaded_dict.py in __setitem__(self, key, value)
    104             self.overloaded[key].set(value)
    105         else:
--> 106             self.data[key] = value
    107 
    108     def __delitem__(self, key):

~/miniconda3/envs/csp/lib/python3.8/site-packages/anndata/_core/views.py in __setitem__(self, idx, value)
     32                 stacklevel=2,
     33             )
---> 34             with self._update() as container:
     35                 container[idx] = value
     36 

~/miniconda3/envs/csp/lib/python3.8/contextlib.py in __enter__(self)
    111         del self.args, self.kwds, self.func
    112         try:
--> 113             return next(self.gen)
    114         except StopIteration:
    115             raise RuntimeError("generator didn't yield") from None

~/miniconda3/envs/csp/lib/python3.8/site-packages/anndata/_core/views.py in _update(self)
     38     def _update(self):
     39         adata_view, attr_name, keys = self._view_args
---> 40         new = adata_view.copy()
     41         attr = getattr(new, attr_name)
     42         container = reduce(lambda d, k: d[k], keys, attr)

~/miniconda3/envs/csp/lib/python3.8/site-packages/anndata/_core/anndata.py in copy(self, filename)
   1526 
   1527             if filename is None:
-> 1528                 raise ValueError(
   1529                     "To copy an AnnData object in backed mode, "
   1530                     "pass a filename: `.copy(filename='myfilename.h5ad')`. "

ValueError: To copy an AnnData object in backed mode, pass a filename: `.copy(filename='myfilename.h5ad')`. To load the object into memory, use `.to_memory()`.

Versions

Scanpy: 1.8.2

Reproduce the failure with the reordered-cell embedding example, then start in scanpy/plotting/_tools/_utils.py at _set_colors_for_categorical_obs and follow its call from scatterplots.py. Done means plotting a categorical color on a reordered backed AnnData view completes without the reported ValueError.