handle cupy array in spread and dynspread functions

[AjayThorve]

This PR aims to fix an issue where cuDF based aggregates when passed to spread & dynspread, throw a typing error.

import datashader as ds
from datashader.transfer_functions import spread, dynspread
import cudf
import numpy as np

df = cudf.DataFrame(np.random.multivariate_normal((0,0), [[0.1, 0.1], [0.1, 1.0]], (1000000,)), columns=['x', 'y'])
dynspread(ds.Canvas().points(df,x='x', y='y')) #or spread(ds.Canvas().points(df,x='x', y='y'))

TypingError: Failed in nopython mode pipeline (step: nopython frontend)
non-precise type pyobject
During: typing of argument at datashader/datashader/transfer_functions/__init__.py (775)

File "../datashader/datashader/transfer_functions/__init__.py", line 775:
def _array_density(arr, float_type, px=1):
    <source elided>
    """
    M, N = arr.shape
    ^

This error may have been caused by the following argument(s):
- argument 0: Cannot determine Numba type of <class 'cupy._core.core.ndarray'>

As per the suggestion made in the corresponding holoviews issue, just converting the aggregate img.data to a numpy from cupy before calling the @nb.jit _array_utils() function seems to fix the issue.

[AjayThorve]

seems like the failing CI is unrelated to this PR, is there anything that can be done to facilitate the merge? would be happy to do it!

[jbednar]

Thanks, @AjayThorve . Agreed; the test failures don't look related. I don't think there's anything for you to do as the author of this PR; the fixes need to be made on master. #1022 might address those, but I've been trying to pay attention to my duties on other libraries and the wheel hasn't spun back around to Datashader just yet. It's overdue for a review of the outstanding PRs and issues and a new release. Sorry about the delay! I'll see if 1022 makes the tests pass, then review the changes there, then come back to this PR...