fix(data_classes): invalidate CogniteResourceList lookups on mutation

[jaideeppyne]

Description

Fixes #1188.

CogniteResourceList builds three identifier→item lookup maps as @cached_property — _id_to_item, _external_id_to_item, _instance_id_to_item — which .get(id=.../external_id=.../instance_id=...) reads. The maps are built lazily on the first .get(...) call.

Only extend() was written to keep those cached maps in sync. Every other membership-mutating operation — append, insert, remove, pop, clear, __setitem__, __delitem__, __iadd__ (+=), __imul__ (*=) — is inherited straight from UserList and mutates self.data without invalidating the caches. So once a map is cached:

• get(id=X) after pop/remove/del returns the removed item (should be None);
• get(id=X) after append/insert/+= returns None for an item that is present.

This is exactly the gap the existing # TODO: We inherit a lot from UserList that we don't actually support... comment flags.

Fix

Add _clear_identifier_lookups() — which drops the three cached_property entries from self.__dict__ so they rebuild on next access — and override each mutating method to call it after delegating to super(). The invalidate-then-rebuild approach is simpler and less error-prone than hand-patching the dicts.

Test

test_get_is_invalidated_by_mutations primes the lookup cache with a get(), then runs each mutator (pop/remove/append/insert/+=/__setitem__/del/clear) and asserts get() reflects the new membership (removed → None, added → found). It fails on the current code (get(id=2) returns the popped item) and passes with the fix.

I ran the TestCogniteResourceList suite: 22 passed. The one remaining failure (test_to_pandas) is a pre-existing pandas-version dtype mismatch unrelated to this change (tracked separately). ruff check/format clean; mypy clean on the changed file.

[gemini-code-assist]

Code Review

This pull request ensures that cached identifier lookup maps in CogniteResourceList are properly invalidated upon mutation by overriding list-mutating methods and adding corresponding unit tests. The review feedback recommends delegating __iadd__ to extend to prevent bypassing duplicate checks, adding a test assertion for this duplicate validation, and replacing Any type hints with stronger types in __setitem__ and __delitem__ to comply with the repository's style guide.

[jaideeppyne]

Thanks for the review — addressed all three in 3d5f563:

• __iadd__ now delegates to self.extend(other), so += enforces the same duplicate-id ValueError (and cache maintenance) as extend() instead of bypassing it via UserList.
• __setitem__ / __delitem__ now use SupportsIndex | slice (and a precise item type) instead of Any.
• Added an assertion that += with a duplicate id raises ValueError.

mypy/ruff clean and the mutation test still passes.

[jaideeppyne]

Small follow-up in 423372e: dropped an unused assignment code from the # type: ignore[index, assignment] on __setitem__. Since mypy.ini sets warn_unused_ignores = true, the redundant code tripped [unused-ignore]; # type: ignore[index] alone is sufficient. mypy and ruff are now clean on the file. Thanks again for the thorough review!