Failing tests on Windows

[sbfnk]

A test is currently failing on Windows. It is not clear what has caused this, and it does not seem to be linked to the latest PR #416 (as it also happens without its main change): https://github.com/epiforecasts/EpiNow2/actions/runs/5554848587/jobs/10147747916?pr=419

[seabbs]

Note that we saw some instability in performance when updating the delay handling code that could be related to this.

[sbfnk]

This now also happens on older version of the code suggesting that it has to do with windows-specific updates to the GHA rather than any specific code updates.

[seabbs]

well that is good news and also I suppose bad news

[seabbs]

I guess for now changing the estimate_secondary test to be a little flimsy would be a good idea

[sbfnk]

I'm not completely convinced the problem is in estimate_secondary as the fitting failures occur on all platforms. Something strange is happening sometimes on windows which may be unrelated
https://github.com/epiforecasts/EpiNow2/actions/runs/5577074428/jobs/10190215287?pr=418#step:6:554

[sbfnk]

The windows failures seem to occur at random times during the testing. Here's another one
https://github.com/epiforecasts/EpiNow2/actions/runs/5584609269/jobs/10207381065#step:6:501

[sbfnk]

I wonder if this could be linked to the recent update to StanHeaders (on CRAN as of 2023-06-14) https://cran.r-project.org/web/packages/StanHeaders/index.html
Noting that Ubuntu actions all seem to be failing more often than they used to with random seg faults.

[seabbs]

That seems very possible though far from ideal

[sbfnk]

In summary, this seems to happen intermittently and at various stages of testing (#418) on current as well as previous versions of EpiNow2 (#421) and with current and older versions of StanHeaders (#427). It's not clear what distinguishes these runs from the ones running successfully on macOS.

[sbfnk]

This seems to have resolved itself without any code change.