Implement grid based merging algorithm (#4820)

* initial implementation of the grid based merging routine

* add compiler directives for different dimensions

* avoid implicit capture of `this` in lambda

* code clean-up

* use `PIdx::x` for r-coordinate while 4667 is still being reviewed

* fix clang-tidy errors

* another clang-tidy fix

* improve doc-strings

* use iterative heap sort since "SYCL kernel cannot call a recursive function"

* fix clang-tidy error

* fix sorting bug; add documentation and CI test

* add dummy CI benchmark file to get proper values

* update benchmark values based on Azure results

* rename `GridBasedMerging` -> `VelocityCoincidenceThinning`

* use `Algorithms::KineticEnergy`

* reorganize merging loop

* update CI benchmark values

* Relativistic correction in product particles' velocity calculation

* update benchmark values after changing energy calculation

* handle edge case with zero cluster momentum

* call redistribute after particle resampling to remove invalid particles

* use unsigned ints for indexing

* Revert "use unsigned ints for indexing"

This reverts commit abe027f3402cf0e7c51c4599a0a50b9e8a6b817a.

* call `Redistribute` before merging

* code clean-up

* also check for `std::isnan` in edge case handling

* add reference for grid based merging

* check that cluster has total weight > 0 before merging

* add defense against numerical error leading to nan

* make resampling message more verbose

* use `deleteInvalidParticles` instead of `Redistribute`

* remove default values for merging parameters

* remove doc-string default specifications in picmi.py

* apply suggestions from code review

* update benchmark values; avoid possible nans

* add assert to prevent merging of photons

* add `BackwardCompatibility` check to `LevelingThinning`