Homogeneously schedule P2P's unpack tasks

[hendrikmakait]

Before this PR, P2P would schedule all unpack tasks on a single worker. For large-scale shuffles/rechunks, this caused the workload to stall and the scheduler to become severely CPU-pressured.

• 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.

    25 files  ± 0      25 suites  ±0   10h 19m 35s ⏱️ + 6m 28s
 4 124 tests  -  1   4 007 ✅  -  3    111 💤 ±0  6 ❌ +2 
47 640 runs   - 11  45 529 ✅  - 13  2 105 💤 ±0  6 ❌ +2 

For more details on these failures, see this check.

Results for commit 1c0a178. ± Comparison against base commit ec3f4ec.

This pull request removes 2 and adds 1 tests. Note that renamed tests count towards both.

distributed.shuffle.tests.test_shuffle ‑ test_unpack_is_non_rootish
distributed.tests.test_scheduler ‑ test_override_is_rootish

distributed.shuffle.tests.test_rechunk ‑ test_homogeneously_schedule_unpack

[jrbourbeau]

Nice! I'm looking forward to taking this for a spin later today

[hendrikmakait]

A/B testing shows no impact (which isn't surprising given that we don't run a good test for this particular case)

(https://github.com/coiled/benchmarks/actions/runs/10851966565)