Different infection estimates between version 1.6.1 and 1.7.0

[avallecam]

Summary:
Getting again a similar issue diagnosed in #829 and solved in version 1.6.1

Description:
In the latest CRAN version, we get again a similar issue that was solved in previous patch version that solved it 1.6.1.

Reproducible Steps:

Using the first code chunk at https://epiverse-trace.github.io/howto/analyses/reconstruct_transmission/estimate_infections.html

Installing using pak::pak("EpiNow2")

packageVersion("EpiNow2")
#> [1] ‘1.7.0’

Installing using pak::pak("[email protected]")

packageVersion("EpiNow2")
#> [1] ‘1.6.1’

---

[sbfnk]

Yes, we reverted the Gaussian Process lengthscale prior to its previous version given issues described in #856 - this will be revisisted in #855 (which is still open) but for now what we have is what works best for the default model. More generally it's tricky for us to find a setting that works for all models so some user tweaking may always be needed.

You can restore the behaviour from version 1.6.1 using gp = gp_opts(alpha = Normal(0, 0.05)).