[codex] Ignore explicit global_phase instructions during separation

[TSS99]

Summary

• ignore explicit global_phase instructions when assigning operations to circuit partitions
• add regression tests for separate_circuit() and partition_problem()

Root cause

Qiskit can materialize a circuit-level global phase as a zero-qubit global_phase instruction. separate_circuit() assumed every instruction touched exactly one partition, so these zero-qubit instructions left partitions_spanned empty and triggered an assertion.

Fix

• skip explicit global_phase instructions while partitioning operations into subcircuits
• keep the change local to _separate_instructions_by_partition() so the public APIs inherit the fix without broader refactoring

Tests

• tox -e lint
• tox -e py311 -- test\\utils\\test_transforms.py test\\test_cutting_decomposition.py test\\test_cutting_workflows.py

Notes or limitations

• this patch intentionally handles the explicit global_phase instruction reported in Global phase circuit instruction causes partition_problem to error. #762
• existing repo-wide pylint config warnings about no-self-use remain unchanged

Fixes #762

[TSS99]

Closing in favor of #817, a cleaner version with a more idiomatic regression test (using GlobalPhaseGate).