fix(graph): reject mismatched collector items for union collection inputs

[unmarco]

Summary

A Collect node connected to an input typed Union[list[X], X] or Optional[list[X]] with mismatched items crashed graph validation with TypeError: issubclass() argument 2 cannot be a parameterized generic, so enqueue returned 500 (which webv2 retries indefinitely) instead of 422.

Collector output validation now reads the destination's item types with the existing extract_collection_item_types(), which handles list branches inside unions, and only calls issubclass() on real classes. Mismatched items are rejected with the usual "matching type" error; matching items are accepted as before.

Before: integer, integer → collect → metadata.items → TypeError → HTTP 500
After: InvalidEdgeError: Collector outputs must connect to a collection input with a matching type → HTTP 422

Related Issues / Discussions

Closes #9852

QA Instructions

• uv run --no-sync pytest tests/test_node_graph.py -k "union_input or optional_list_input or list_first_union": the two new rejection cases fail on main with the TypeError and pass with this change; the two acceptance cases pass on both.
• uv run --no-sync pytest tests/test_node_graph.py tests/test_graph_execution_state.py tests/app/services/shared tests/app/services/test_workflow_graph_builder.py tests/app/routers/test_session_queue_sanitization.py -n 4: 1753 passed, 22 skipped, 2 xfailed.
• uv tool run ruff@0.11.2 check / format --check on the changed files: clean.
• Batch.model_validate on the repro graph from the issue: ValidationError (422) instead of TypeError (500). Also checked merge_metadata.collection and core_metadata.loras: both now raise InvalidEdgeError.
• Not run: the full pytest -n logical suite, and no browser check (no UI change).

Review

No blocking findings. Intended behaviour changes beyond the crash, all toward reading the annotation correctly:

• Union[int, list[str]] with integer items was accepted, because the scalar happened to come first. It is now rejected.
• Union[list[A], list[B]] matches if any list branch matches, not only the first.
• Optional[list[Any]] / Union[list[Any], X] accept collected items, typed or untyped, as list[Any] already did. Before, they crashed or were rejected. No core node uses such an annotation today.

Checklist

• The PR has a short but descriptive title, suitable for a changelog
• Meaningful regression coverage added / updated where needed; obsolete tests/code removed
• Persisted-state and API changes include required migrations / compatibility validation
• Relevant performance/efficiency opportunities considered; material claims have evidence
• Material review findings resolved and relevant checks rerun
• Documentation added / updated (if applicable)
• Updated What's New copy (if doing a release after this PR)

---

^{Need help on this PR? Tag @codesmith-bot with what you need.}

[JPPhoto]

Hi @unmarco, thanks for the PR.

In my review, I uncovered this:

• invokeai/app/services/shared/graph_validation.py:2340: For a collector of Derived(Base) targeting list[Union[Base, str]], the union-aware subtype check misses the subclass, and the class guard skips issubclass; validation rejects a valid edge. The base accepted it. Effect: valid workflows cannot connect this collector. Likelihood: plausible when a node declares union-typed list items. Recovery: manual graph or type workaround. Test: the candidate validator reproduces the rejection; add a Graph.add_edge regression for this case.

Please add the test and fix the code and perform another self-review, then re-request review from me when complete.

[unmarco]

Thanks for catching this. Fixed in 796b0f8.

_collector_item_matches now checks issubclass against each member when the item type is a union (typing.Union or PEP 604 X | Y). A Derived(Base) collector is accepted by list[Union[Base, str]] again, and mismatched items are still rejected.

I added Graph.add_edge regressions for both the accept and reject cases, each run with both union spellings. On the previous commit, both accept cases fail.

Self-review and checks:

• tests/test_node_graph.py and tests/test_nodes.py pass (158 tests), and ruff check / ruff format --check are clean.
• The graph engine, session queue and shared-service suites pass (1697 passed, 15 skipped, 2 xfailed).
• End to end, the original repro (Integer ×2 → Collect → Metadata items) now returns 422 from enqueue_batch instead of the 500 on 7.0.0-alpha.2. A valid MetadataItem ×2 → Collect → Metadata graph still enqueues and completes.