Migrate dependencies dependents sets from TaskState to Key

[fjetter]

This is a follow up to #9036 and builds directly on top

This replaces the elements in TaskState.dependencies with the keys instead of the TaskState objects themselves. This has a few benefits

1. It is a step towards making TaskState overhead lighter. A few months ago we were investigating GC overhead on the scheduler and the TaskState objects are adding to this in a non-trivial way. The fewer references we keep around the easier. Besides, these dependencies/dependents links are cycles that have to be followed by the GC. I don't expect this change to have a measurable impact but I believe this direction is healthy for the scheduler as a whole.
2. More importantly, this paves the way to drop the dedicated dependencies set altogether in favor of the frozenset of the Task class (TaskState.run_spec) and reduce update_graph runtime and reduces memory overhead a little (c.f. a simple, empty set needs already 216B on py3.10, see also Reduce memory usage of scheduler process - Optimize scheduler.py::TaskState class #8331)

[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   12h 51m 31s ⏱️ + 1h 38m 14s
 4 101 tests + 1   3 951 ✅  -  33    112 💤 ±0   38 ❌ + 34 
51 413 runs  +13  48 802 ✅  - 285  2 308 💤  - 1  303 ❌ +299 

For more details on these failures, see this check.

Results for commit e72e219. ± Comparison against base commit 16aa189.