Remove GPU executor

[hendrikmakait]

Closes #8338 by removing the GPU executor. I am not aware of any usage of the GPU executor, so I'd recommend ripping it out instead of worrying about maintenance. I'm happy to reconsider if anybody brings an example of the executor being used to my attention.

• Tests added / passed
• Passes pre-commit run --all-files

[github-actions]

Unit Test Results

See test report for an extended history of previous test failures. This is useful for diagnosing flaky tests.

       27 files  ±  0         27 suites  ±0   11h 51m 24s ⏱️ - 2m 59s
  3 937 tests  -   1    3 824 ✔️  -   2     110 💤 ±0    3 ❌ +1 
49 523 runs   - 13  47 221 ✔️  - 14  2 290 💤 ±0  12 ❌ +1 

For more details on these failures, see this check.

Results for commit 8e4c6d2. ± Comparison against base commit 5d9e406.

This pull request removes 1 test.

distributed.tests.test_worker ‑ test_gpu_executor

[hendrikmakait]

cc @dask/gpu

[pentschev]

No objections from me, I'm unaware of any uses of that in production code.

[jakirkham]

Think this is something Matt added a while back ( #5084 ). There is some more context in issue ( #5084 ) in particular this comment ( #5084 (comment) )

IIUC it was meant as an experiment/proof-of-concept that would be followed up with other work. While there was related work (annotations), think that pivoted into HLGs, etc. So don't think the original idea went anywhere

If someone were to explore a GPU Executor today, probably a better answer would be add it to Dask-CUDA and make the minimal changes to Distributed (if any) to allow it to be plugged in. This might be a good approach to consider for other CUDA bits

All this to say ripping the GPU Executor out sounds like the right approach

[fjetter]

Thanks everyone for chiming in!