Add support for missing NAs in estimate_infection() model

[seabbs]

This PR adds support for missing data in the form of NAs but dropping them from the likelihood rather than imputing that they are zero and then optionally

Note that this is a breaking change and may be undesirable for some users where missing dates do indicate zeros vs being truly missing.

I have tested by knocking out data from the example and running the model.

[github-actions]

This is how benchmark results would change (along with a 95% confidence interval in relative change) if cb5705c is merged into main:

•   :ballot_box_with_check:default: 32.5s -> 32.2s [-18.26%, +16.05%]
•   :ballot_box_with_check:no_delays: 33.5s -> 32.3s [-19.32%, +12.51%]
•   :ballot_box_with_check:random_walk: 9.78s -> 9.97s [-5.78%, +9.75%]
•   :ballot_box_with_check:stationary: 1.18m -> 20.6s [-232.86%, +91.02%]
•   :ballot_box_with_check:uncertain: 50.1s -> 53.3s [-5.13%, +17.82%]
  Further explanation regarding interpretation and methodology can be found in the documentation.

[sbfnk]

Very nice! A couple of minor questions/confusions on my side.

[github-actions]

This is how benchmark results would change (along with a 95% confidence interval in relative change) if 876a533 is merged into main:

• ❗🐌default: 33.6s -> 37.9s [+0.54%, +24.72%]
•   :ballot_box_with_check:no_delays: 34.8s -> 37.3s [-9.79%, +24.19%]
•   :ballot_box_with_check:random_walk: 10.5s -> 9.47s [-32.17%, +13.29%]
•   :ballot_box_with_check:stationary: 20s -> 19.3s [-22.76%, +16.34%]
•   :ballot_box_with_check:uncertain: 51s -> 48.5s [-25.08%, +15.29%]
  Further explanation regarding interpretation and methodology can be found in the documentation.