Enh: Defer expensive loading of dask.array

[psobolewskiPhD]

• Closes Lazy import of compat layer optional dependencies #2265
  • this isn't a comprehensive solution for every array type, but addresses the big one dask.array
• Executed pre-commit run --all-files or pixi run lint --all-files with no errors
  • I ran linting manually, don't have pixi handy
• The change is fully covered by automated unit tests
• Documented in docs/ as appropriate
  • NA?
• Added an entry to the CHANGES file
  • will do so if this seems like something y'all are on board with

This is a performance enhancement, aimed at the import time of pint.
As noted int he issue, dask.array is a expensive import, so in this PR I attempt to address the issue by deferring it's loading from import pint (dask/dask.base in contrast are pretty light, so I leave them alone to keep this simple.)
With this PR (all are second runs)

• python -X importtime -c 'import pint' bottom line:
  • import time: 649 | 176212 | pint
• python -X importtime -c 'import pint' 2>&1 | grep dask.array
  • nothing
• python -X importtime -c 'import pint' 2>&1 | grep dask
  • shows dask.base as 31 ms on my machine

With main:

• python -X importtime -c 'import pint' bottom line:
  • import time: 630 | 465967 | pint
• python -X importtime -c 'import pint' 2>&1 | grep dask.array
  • import time: 244 | 278716 | dask.array
• python -X importtime -c 'import pint' 2>&1 | grep dask.base
  • shows dask as 31 ms

With this PR in pint we can tuck some import dask.array as da imports in napari into functions and save ~250 ms on our startup time.

Edit: this has a nice side-effect of also resulting in import pint not importing pandas.

[andrewgsavage]

looks good, do add a changes note

[andrewgsavage]

doctest fail looks unrelated

[codspeed-hq]

Merging this PR will not alter performance

✅ 448 untouched benchmarks

---

_{Comparing psobolewskiPhD:lazy_dask (fbfe9bc) with master (86f37c0)}

[psobolewskiPhD]

looks good, do add a changes note

Thanks! Done! I wasn't sure if I should reference the issue or the PR in the parens. I went with the issue number?
Happy to edit it or please feel free to push.

[psobolewskiPhD]

Oh interesting, it looks like dask.array was also pulling in pandas.
With this branch:
python -X importtime -c 'import pint' 2>&1 | grep pandas

• nothing
  With main:
  import time: 183 | 130141 | pandas

[psobolewskiPhD]

The lint fail appears to be due to pixi version?
https://github.com/hgrecco/pint/actions/runs/22461640461/job/65057385529?pr=2276#step:4:6
Do you want me to try to fix that here? or another PR?
I'm not a pixi user, but I can read docs and give it a shot -- the error seems clear enough, regarding the change from project to workspace