Refining FLArE outputs

#16 is making the first step for restructuring FLArE output trees. Currently we have:
- `flare_hits`: full hits from the liquid argon volume;
- `hcal_hits`: full hits from the HCAL (BabyMind or legacy);
- `.h5` file with pixel map 3D output;
- (optional) 2D histograms, per event and per prong;
- (optiocal) pseudo-reco tree;

There a few points to think about:

1. The size of `flare_hits` can be challenging. I did a series of quick tests using the same 5 genie events. The current solution is to remove the step size user limit. Is this acceptable?
Can we reduce the output size by merging hits in 3D? For instance, we could do something similar to what `PixelMap3D` does at a single particle level instead of binning all energy deposits from all particles together.
But if we do so, isn't it the same of having larger steps? Should we just settle to larger step sizes?
```
no max step size -> ~20 MB/event
1.0 mm max step size -> ~33 MB/event
0.5 mm max step size -> 64 MB/event
```

2. Enabling the pseudo-reco output requires a lot of acrobatics... and it's currently no longer working as before because we are not updating the primaries list (only using native G4 vertex primaries). There are also places in which we require a `FPFNeutrino`, which can't always be there.
Given the proposed changes in #18 -- full particle hierarchy tree in output -- can we completely move the pseudo-reco in a downstream ad-hoc FLArE-only package?

3. Should create separate hit/SD classes for the FLArE HCAL?