Generalize along-step kernels

[sethrj]

Currently we use the same along-step kernel (including slowing down and field propagation) for all particle types, all energies, all regions. It's probably silly to use the field propagator to move a neutral particle (but at least it works!), we'll likely need to have different slowing-down models for different energy regimes, and we need to support more extensible per-region fields and settings. (Neutral vs "user" action is supported as of #745.)

We will partition the multi-D {particle, energy, region} space into blocks of AlongStepIds, each of which has an MscId, PropagationId, and ElossId. This will allow users to define easy host-side schemes to select a kernel (e.g., looping over particle types and setting different kernels for charged and neutral).

Execution plan

• [ ] Define Celeritas regions, replacing GeoMaterialData #983
• [ ] Import regions (and build mag field regions) from Geant4
• [ ] Build ragged region+particle+energy mapper class (note: may use in future for selecting physics model?)
• [ ] Add along-step parent class with basic interface, for now taking existing along-step actions and (during execute) setting along-step action IDs (replacing the last bit of PreStepExecutor)
• [ ] Remove volume-specific field from uniform propagator
• [ ] Split along-step actions into propagator/eloss/msc with kernels in between

Current kernels

Kernel	CylMapFieldMsc	RZMapFieldMsc	UniformMsc	GeneralLinear	Neutral
limit-step-msc-urban	✓[^]	✓*	✓*	✓*
propagate-cylmap	✓
propagate-rzmap		✓
propagate-uniform			✓
propagate-linear			✓*	✓	✓†[^†]
scatter-msc-urban	✓*	✓*	✓*	✓*
update-time	✓	✓	✓	✓	✓†
apply-eloss-fluct	✓*	✓*	✓*	✓*
apply-eloss-mean	✓*	✓*	✓*	✓*
update-track	✓	✓	✓	✓	✓†

[^*]: * Conditional
[^†]: † Combined into a single kernel on GPU

[sethrj]

Note that we should probably add regions (set of volumes) instead of materials; and then refactor our cutoff/materials so that materials can be reused across regions with different cutoffs.

[sethrj]

@amandalund This is what I had in mind for having separate MSC methods as a function of energy. With the speedup we saw from breaking the along-step kernels in to smaller kernels, though, we may want to group tracks by propagation sub-kernel first, and then have a secondary set of kernels for the different MSC regimes.