✨ Expand Qiskit RL actions with stochastic selection

[flowerthrower]

🤖 AI text below 🤖

Description

Expands the RL action space with focused Qiskit layout, routing, and optimization passes: TrivialLayout, ElidePermutations, SabreSwap, BasicSwap, LookaheadSwap, RemoveIdentityEquivalent, and Optimize1qGatesSimpleCommutation.

ElidePermutations preserves its virtual permutation through the canonical layout, and OptimizeCliffords collects Clifford blocks before optimizing them. Existing RL models must be retrained because the action schema changes.

QiskitSabreMapping and SabreSwap are stochastic. They evaluate 20 candidates by default and retain the best configured figure of merit after target-basis translation. Supported scoring failures fall back to minimizing inserted SWAPs; failed candidates do not abort remaining attempts, and TimeoutError is propagated. Set stochastic_action_trials=1 for single-attempt selection.

IBM AI routing remains excluded because the published packages conflict with this stack's NetworkX requirement and import Qiskit internals removed from the resolved Qiskit version.

This is position 3 of the stack. It depends on #758 and is followed by #786.

Fixes #667

Part of #675

Checklist

• The diff is focused and follows repository conventions.
• AI assistance was authorized and is disclosed.
• This body begins with the required visible disclosure.
• I confirm that I have personally reviewed and understood all AI-generated content.

[flowerthrower]

🤖 AI text below 🤖\n\nSuperseded while rebuilding the corrected downstream stack after #758.