Fix timings in tests

[shikokuchuo]

Fixes #236.

The failing tests have not been touched for multiple years, nor has code that affect these been changed in this release.
Also we make extensive rhub check runs prior to release (including MKL) e.g. at https://github.com/r-lib/later/actions/runs/17038400115.

I am able to re-produce the series of 3 test failures if and only if the run_now() in the g() function fails to execute the prior later() call.

• Reproduce by removing the 0.5 delay in later() and commenting out the run_now() in g().
• This causes the cascade of subsequent run_now() calls to run the previous later call rather than the expected one
• If the initial test actually failed, we would get different error messages for the subsequent errors (if they failed independently themselves)

Although implausible, seems not impossible under extreme load.

The solution is to simply wait a full 2 secs in g().