✨ Add Qiskit and stochastic RL actions

[flowerthrower]

🤖 AI text below 🤖

Description

Creates the first downstream layer after the reviewed #758 observation PR.

This layer contains the remaining compact scalar observation encoding together with the Qiskit action and stochastic-selection changes:

• encode num_qubits and depth as normalized one-element float32 arrays;
• add TrivialLayout, ElidePermutations, SabreSwap, BasicSwap, LookaheadSwap, GateDirection, RemoveIdentityEquivalent, and Optimize1qGatesSimpleCommutation actions;
• collect Clifford blocks before OptimizeCliffords and decompose its result;
• evaluate 20 candidates by default for actions marked stochastic and retain the best configured figure of merit;
• continue after individual failed candidates while propagating pass and scoring timeouts immediately.

Existing RL models must be retrained because both the observation and action schemas change. The IBM AI routing actions remain excluded because the published IBM transpiler packages conflict with the dependency versions required by this stack.

This draft depends directly on #758.

Validation

• repository lint;
• focused observation, routing, mapping, optimization, and stochastic-action checks;
• direct pass- and scoring-timeout smokes.

Fixes #667

Part of #675

Part of #664

Checklist

• The pull request contains only the changes assigned to this stack layer.
• Existing focused tests and release metadata cover the changed behavior.
• The changes follow the repository style and introduce no new warnings.
• I reviewed the complete base-to-head diff.

If PR contains AI-assisted content:

• Any agent that created, edited, or submitted GitHub content was explicitly authorized for that scope.
• Every agent-authored or agent-edited public text body begins with the visible disclosure 🤖 *AI text below* 🤖.
• I confirm that I have personally reviewed and understood all AI-generated content, and accept full responsibility for it.

[flowerthrower]

🤖 AI text below 🤖

Recreated after restoring #758 to its human-reviewed commit, so this temporary downstream diff is retired.