Dask Concatenation Should Impute With Dask Array, not Numpy

[ilan-gold]

Please make sure these conditions are met

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

Report

Code:

import anndata as ad
import dask.array
shape = (10, 100)
adata1 = ad.AnnData(layers={'foo': da.random.random(shape)})
adata2 = ad.AnnData(X=da.random.random(shape))
type(ad.concat([adata1, adata2], join="outer").layers['foo'])

Internally this will call missing_element which can be improved as a comment states:

anndata/src/anndata/_core/merge.py

Lines 962 to 968 in 997fbd7

	# Handling of missing values here is hacky for dataframes
	# We should probably just handle missing elements for all types
	result[k] = concat_arrays(
	[
	el if not_missing(el) else missing_element(n, axis=axis)
	for el, n in zip(els, ns)
	],

Versions

-----
anndata             0.11.1.dev12+gd61e09c8
session_info        1.0.0
-----
asciitree           NA
cloudpickle         3.1.0
cython_runtime      NA
dask                2024.11.2
dateutil            2.9.0.post0
exceptiongroup      1.2.2
h5py                3.12.1
importlib_metadata  NA
natsort             8.4.0
numcodecs           0.13.1
numpy               2.1.3
packaging           24.2
pandas              2.2.3
psutil              6.1.0
pytz                2024.2
scipy               1.14.1
six                 1.16.0
tlz                 1.0.0
toolz               1.0.0
yaml                6.0.2
zarr                2.18.3
zipp                NA
zoneinfo            NA
-----
Python 3.10.14 | packaged by conda-forge | (main, Mar 20 2024, 12:45:18) [GCC 12.3.0]
Linux-5.14.0-427.37.1.el9_4.x86_64-x86_64-with-glibc2.34
-----
Session information updated at 2024-11-22 18:06