Fix non-deterministic rewriter behavior in multi-output pattern matching

[Copilot]

GraphPattern.__init__ collects output nodes into a set[NodePattern], then converts to a list via list(output_nodes). Python's hash randomization means this list has non-deterministic order across process invocations. For multi-output patterns (e.g. SlicesSplit), the matcher fixes output_nodes[0] to the current node and searches for output_nodes[1:] — so different orderings cause the same rule to match or not match non-deterministically.

• Replace set[NodePattern] with dict[NodePattern, None] (ordered set idiom) to preserve deterministic insertion order from the outputs sequence
• Add regression test verifying output_nodes ordering is stable across repeated GraphPattern constructions

# Before: non-deterministic iteration order
output_nodes: set[NodePattern] = set()
...
output_nodes.add(candidate)
...
self.output_nodes: list[NodePattern] = list(output_nodes)

# After: preserves insertion order from outputs sequence
output_nodes: dict[NodePattern, None] = {}
...
output_nodes[candidate] = None
...
self.output_nodes: list[NodePattern] = list(output_nodes)

Fix #2878

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.
✅ Project coverage is 72.04%. Comparing base (1077da7) to head (7be2601).
⚠️ Report is 2 commits behind head on main.
✅ All tests successful. No failed tests found.

Additional details and impacted files

@@            Coverage Diff             @@
##             main    #2880      +/-   ##
==========================================
+ Coverage   71.96%   72.04%   +0.07%     
==========================================
  Files         239      239              
  Lines       29224    29305      +81     
  Branches     2878     2880       +2     
==========================================
+ Hits        21031    21112      +81     
  Misses       7216     7216              
  Partials      977      977              

☔ View full report in Codecov by Sentry.
📢 Have feedback on the report? Share it here.