WIP: add a way to do reprojection without dumping dask arrays to memory maps

[astrofrog]

Very much just work in progress, not ready for testing

[codecov[bot]]

Codecov Report

:x: Patch coverage is 98.07692% with 1 line in your changes missing coverage. Please review. :white_check_mark: Project coverage is 91.06%. Comparing base (4f8e9eb) to head (939393b). :warning: Report is 16 commits behind head on main.

Files with missing lines	Patch %	Lines
reproject/array_utils.py	97.61%	1 Missing :warning:

Additional details and impacted files

@@            Coverage Diff             @@
##             main     #532      +/-   ##
==========================================
+ Coverage   90.12%   91.06%   +0.94%     
==========================================
  Files          30       30              
  Lines        1822     1858      +36     
==========================================
+ Hits         1642     1692      +50     
+ Misses        180      166      -14     

:umbrella: View full report in Codecov by Sentry.
:loudspeaker: Have feedback on the report? Share it here.

:rocket: New features to boost your workflow:

• :snowflake: Test Analytics: Detect flaky tests, report on failures, and find test suite problems.

[astrofrog]

This seems to work - the main question now will be whether to rename the 'none' option or change the keyword argument to memmap_dask=True/False for clarity.

[astrofrog]

@Cadair @svank this adds a dependency on a very recent version of dask-image, which means also bumping the minimum dask version to one that is around a year old. I don't really foresee issues with this since pip/conda should just do their thing when people install reproject, but let me know if you are opposed to adding this dependency.

[Cadair]

I'm fine with that, very happy for dask-native reproject!