Refactor AdePT into 2 libraries 5: Refactor G4HepEm data management, unify initialization

[SeverinDiederichs]

This PR belongs to the refactor of AdePT described in #516.
This PR is baed on #521 and should not be reviewed before that one is merged.

This PR moves the G4HepEm host-data preparation out of AsyncAdePTTransport into a dedicated host-side helper and unifies the initialization.

Before the initialization was:

1. Initialization of AdePTTransport. This created the G4HepEmState and uploaded already to the GPU
2. Initialization of VecGeom and InitVolAuxData on CPU, using the G4HepEmState.
3. Initialization of the rest on the GPU.

With this PR, the G4HepEm initialization is factored out of the AdePTTransport initialization.
This allows for a cleaner workflow:

1. Initialize all the host data. The CheckGeometry and InitVolAuxData functions use the G4HepEmData directly that was prepared on the host
2. Upload the host-initialized data to the GPU, which includes the G4HepEmData as well as geometry and InitVolAuxArray.

Until now, the transport rebuilt the HepEm host data internally and then also used it to steer the remaining host-side initialization. This made the initialization flow harder to follow and mixed host preparation with device-side transport setup.

The G4HepEm data is now factored out into a HepEmHostData class, which takes care of the initialization and life-time management.

It was verified that this PR

• Changes physics results
• Does not change physics results

[phsft-bot]

Can one of the admins verify this patch?

[SeverinDiederichs]

Solved: Drift test with WDT failed, this needs to be investigated first.

The issue was the ownership of the G4HepEmParameters: while the G4HepEmData was created freshly for the AdePTTransport, the parameters were only borrowed from the owning G4 worker thread. This could cause a crash in the tear down, as other threads may finish later and it would cause a destruction order bug.

This is fixed by doing a deep copy of the G4HepEmParameters.

That means: each G4 worker owns a G4HepEmTrackingManager, which owns G4HepEmParameters. At the same time, the global AdePTTransport owns its own G4HepEmParameters, which it needs it for the GPU upload.