speed up generation of infections

[sbfnk]

implements suggestions from #415

[github-actions]

This is how benchmark results would change (along with a 95% confidence interval in relative change) if 8929be0 is merged into main:

•   :rocket:default: 57.2s -> 50.2s [-20.5%, -3.69%]
•   :rocket:no_delays: 57.3s -> 50.2s [-22.78%, -1.82%]
•   :ballot_box_with_check:random_walk: 15.6s -> 14.5s [-18.12%, +3.61%]
•   :ballot_box_with_check:stationary: 31.9s -> 30.5s [-12.24%, +3.69%]
•   :rocket:uncertain: 1.34m -> 1.08m [-32.49%, -6.36%]
  Further explanation regarding interpretation and methodology can be found in the documentation.

[github-actions]

This is how benchmark results would change (along with a 95% confidence interval in relative change) if 59dae8e is merged into main:

•   :rocket:default: 55s -> 47.4s [-25.82%, -1.69%]
•   :ballot_box_with_check:no_delays: 56.2s -> 49.4s [-26.33%, +1.98%]
•   :rocket:random_walk: 16.3s -> 14.3s [-21.74%, -2.55%]
•   :rocket:stationary: 30.5s -> 29.4s [-6.44%, -0.37%]
•   :ballot_box_with_check:uncertain: 1.17m -> 1.14m [-14.17%, +9.48%]
  Further explanation regarding interpretation and methodology can be found in the documentation.

[seabbs]

LGTM.

Note this may reduce stability in contexts with problematic data which were the motivation for adding these padding values.

[sbfnk]

I was wondering about that, thinking that it would have been mitigated by cd2e738 (#366) but I might be missing something.

[seabbs]

Potentially though it was largely issues in warm-up from the likelihood vs in the generated quantities. I think given pathfinder will soon be out many of these issues with stability are likely soon to be resolved.