Support >2B Particles per MPI Rank

[ax3l]

Support more than 2 billion particles per rank (process or device).

This gives us a small performance hit (~5%, to double-check), but makes sure one never has to worry. This is useful for nodes or GPUs with very large memory (e.g. >80 GB) when doing very large (>2B particle) runs on them.

• rebase after Reuse Particle Chunking Logic #1072
• measure performance difference
• technically needs a whole lotta updates in AMReX, too.

[codspeed-hq]

Merging this PR will not alter performance

✅ 37 untouched benchmarks

---

_{Comparing ax3l:topic-long (d352c0f) with development (daef9a8)}

[ax3l]

(Emulated) perf on CPU is the same for particle pushes :-o
Fun and unexpected!