Fix PhaseSpaceActor scoring in nested attached volumes

[nkrah]

Summary:

• add a GateVActor helper to detect whether a step occurs in the top attached volume rather than a daughter
• use that helper in GatePhaseSpaceActor to ignore subtree-triggered internal surface crossings

Motivation:
PhaseSpaceActor is attached to a single volume in GATE, but the underlying sensitive detector registration is recursive over the attached volume hierarchy. This can lead to surprising extra phase-space entries when daughter volumes are crossed.

Changes:

• add GateVActor::IsStepInTopAttachedVolume(const G4Step*)
• early-return in GatePhaseSpaceActor::SteppingAction() unless the step is in the top attached volume

Notes:

• this keeps the framework helper isolated from the actor behavior change in two separate commits
• it does not change recursive SD registration globally; it only tightens PhaseSpaceActor behavior

[tbaudier]

Hello @nkrah
I think your PR modify the behavior of the test: geometry/test058_iec_six_spheres.py