fix: make TaskInstance and Realization hashable

[Adityakk9031]

closes:#4

Summary:

TaskInstance and Realization are decorated as frozen dataclasses, but they contain dictionary fields. By default, Python's generated hash method attempts to hash these mutable dictionaries, raising a TypeError. This PR implements custom hash methods for both classes to convert their dictionary fields into sorted tuples of items, resolving the unhashable type error and allowing them to be used in sets or as dictionary keys.

Changes:

• src/kitchenbench/instances.py: Added custom hash implementations to Realization and TaskInstance.
• tests/test_instances.py: Added test_instances_and_realizations_are_hashable to assert that TaskInstance, Realization, and TaskSpec instances are hashable.

Verification:

• Ran all tests via pytest, maintaining 100% code coverage.
• Verified that ruff checks and mypy type checks pass.

[Adityakk9031]

@jeqcho have a look

[jeqcho]

@Adityakk9031 Hi, happy to review again if the failing tests are fixed.

[jeqcho]

Thanks @Adityakk9031!

On the red CI: the macOS py3.12 failure was our infrastructure, not you — setup-uv got rate-limited before anything of yours ran, and I've re-run that job. The lint failure is real but trivial: ruff format --check wants to reformat src/kitchenbench/instances.py and tests/test_instances.py. Run uv run ruff format ., commit, push, and it should go green. I'll review properly once it is.

[Adityakk9031]

@jeqcho have a look

[Adityakk9031]

@jeqcho check latest commit

[jeqcho]

Merged, thanks @Adityakk9031. The part I want to call out: relying on dataclass(frozen=True) keeping an explicitly defined __hash__ rather than fighting it with unsafe_hash or eq=False was the clean way through, and covering TaskSpec in the test even though the PR never touches specs.py shows you traced the transitive fix (it embeds tuple[TaskInstance, ...]). Nothing in either hash is dropped relative to __eq__, which is the invariant that matters.

Review surfaced two small follow-ups, written up as #30: an equal-objects-equal-hash test (the current test passes even under identity hashing), and the static: dict[str, Any] annotation that could let a future spec silently reintroduce the bug this PR fixed. First refusal is yours if you want them. Thanks again for sticking through the format rounds.