instant field ionization transitions

[BrianMarre]

implementation of instant field ionization transitions.

• requires PR bug fix ApplyIPDIonization #5207 to be merged first
• requires PR Fix infinite ADK rates #5206 to be merged first
• requires PR refactor to avoid code duplication #5220 to be merged first
• requires PR Further refactoring of AtomicPhysics to avoid code duplication #5227 to be merged first
• needs to be rebased to dev

Also fixes a bug in the FillRateCache atomicPhysics(FLYonPIC) sub-stage and removes the unused probability estimator implementations, FLYonPIC currently always uses linear probability estimation only.

Only commits after refactor Record Changes kernel are part of this PR, all other commits belong to a base PR.

Attention! This PR breaks setups using custom atomicPhysics.param files. To fix a setup add the following new member to the RateSolverParam struct.

/** maximum number of atomicPhysics sub-steps per PIC time step
 *
 * all rates above timeStepAlpha / maximumNumberSubStepsPerPICTimeStep will be solved in
 *  rate fix point(equilibirium) approximation only.
 *
 * @attention This limit is currently only enforced for field ionization transitions! If the number of
 * sub-steps required by collisional processes is higher than the limit set, AtomicPhysics will perform more
 * subs-steps than set here.
 */
static constexpr uint32_t maximumNumberSubStepsPerPICTimeStep = 200u;

Attention! This PR breaks setups using custom atomicPhysics_Debug.param files. To fix an setup add the following to the old file.

//! @attention performance relevant
namespace picongpu::atomicPhysics::debug::kernel::chooseTransition
{
    //! @attention performance relevant
    constexpr bool CHECK_FOR_INFINITE_FIELD_IONIZATION_RATES = false;
} // namespace picongpu::atomicPhysics::debug::kernel::chooseTransition
namespace picongpu::atomicPhysics::debug::kernel::chooseInstantTransition
{
    //! @attention performance relevant
    constexpr bool CHECK_FOR_INFINITE_FIELD_IONIZATION_RATES = false;
} // namespace picongpu::atomicPhysics::debug::kernel::chooseInstantTransition

[BrianMarre]

@psychocoderHPC the failed CI-jobs got stuck somehow, please restart them

[psychocoderHPC]

I wrote it already in an other PR. It looks very dangerous to use a reference here.
If it is a pmacc box please take the return value always by copy. There is only one case where a reference is ok and this is if you use native alpaka to get shared memory.

[psychocoderHPC]

Could be that this should be change in other code parts too.

[BrianMarre]

The use of a reference is correct and necessary here, since the dataBox will return a reference to global device memory and we don't want to store a copy of the superCell's rejection probability cache for every worker due to its significant size, ~ 512 * 32-bit.

[psychocoderHPC]

short rule: if you write code like type& x = function(); this is in 99% of all cases wrong and requires removing the reference.
This is a typical case to produce segfaults.

[BrianMarre]

A general comment, most AtomicPhysics(FLYonPIC) sub stages still use the old auto& <name> = *dc.get<type>(<ID>) style. I updated all new and all changed sub-stages in this PR but we will have to update all other old stages separately.