Fix cycle error in FusionOptimizer

[jaanerik]

Fix cycle error in FusionOptimizer by preventing overlapping subgraphs and bitset corruption

Description

This PR addresses a structural regression in FusionOptimizer (introduced in recent ordering changes) where certain complex topologies or multi-output subgraphs could trigger a ValueError: graph contains cycles during the final topological sort of discovered subgraphs.

The Problem:

1. Overlapping Subgraphs: The discovery loop did not check if a node was already "claimed" by a previously discovered subgraph. This allowed nodes to be shared between subgraphs, creating circular dependencies in the meta-graph of subgraphs.
2. Bitset Corruption: When a multi-output subgraph was discovered, the logic for updating successor dependencies incorrectly updated nodes within that same subgraph. This caused ancestors to appear as if they depended on their own descendants, corrupting the priority keys used for expansion and leading to non-convex subgraphs.

Related Issue

• Closes BUG: walk_toposort spuriously raises 'graph contains cycles' on acyclic int-keyed graphs > 256 nodes #2170

Checklist

• Checked that the pre-commit linting/style checks pass
• Included tests that prove the fix is effective or that the new feature works
• Added necessary documentation (docstrings and/or example notebooks)
• If you are a pro: each commit corresponds to a relevant logical change

Type of change

• Bug fix

[ricardoV94]

I can't reproduce the issue and the tests were passing fine on main. Closing this, feel free to comment on the original issue if you have a working failing example