[BUG] support float64 .X for cell selection from gene expression histogram.

Copying the comment I made here as a fresh issue.

Using the demo data at https://cellxgene-example-data.czi.technology/pbmc3k.h5ad if I run

import scanpy as sc
import numpy as np
import scipy
adata = sc.read('pbmc3k.h5ad')
adata.X = scipy.sparse.csr_matrix(adata.X, dtype=np.float32)
adata.write('pbmc3k_32.h5ad')

adata = sc.read('pbmc3k.h5ad')
adata.X = scipy.sparse.csr_matrix(adata.X, dtype=np.float64)
adata.write('pbmc3k_64.h5ad')

Then the highlighting and selection as a group of cells by selecting a region of the gene expression histogram only works for pbmc3k_32.h5ad in cellxgene v0.17.0

This is a problem because h5ad files produced by SeuratDisk seem to use float.64

Which is probably also the cause of https://github.com/chanzuckerberg/cellxgene/issues/2291 since the Azimuth data h5ad file was likely produced with SeuratDisk and this file looks like:

adata = sc.read('local.h5ad')
adata.X
<982538x77521 sparse matrix of type '<class 'numpy.float64'>'
	with 1748651019 stored elements in Compressed Sparse Row format>

Screenshots

Expected behavior cellxgene supports this function for float64 .X matrix

[Jon-bioinfo]

I see the same error.

crossfilter.js:279 Warning - unknown metadata schema (float64) for field X 1480.
crossfilter.js:247 Uncaught (in promise) TypeError: o is not iterable (cannot read property undefined)
    at AM._addObsCrossfilterDimension (crossfilter.js:247)
    at crossfilter.js:166
    at Generator.next (<anonymous>)
    at Cr (asyncToGenerator.js:3)
    at i (asyncToGenerator.js:25)

[bkmartinjr]

@Jon-bioinfo - thanks for the issue.

Could you take a look at the Browser console and report any error messages you see there?

FYI, I am able to reproduce, and the console shows:

Warning - unknown metadata schema (float64) for field X 1278.
_getObsDimensionParams @ crossfilter.js:279
crossfilter.js:247 Uncaught (in promise) TypeError: o is not iterable (cannot read property undefined)
    at AM._addObsCrossfilterDimension (crossfilter.js:247)
    at crossfilter.js:166
    at Generator.next (<anonymous>)
    at Cr (asyncToGenerator.js:3)
    at i (asyncToGenerator.js:25)

Likely scenario: the back-end is failing to down-cast to float32 correctly (the front-end only accepts float32).