SI.MonadDelay.threadDelay behaves differently than in IO for sub-microsecond timeouts

[isovector]

Describe the bug

The SI.MonadDelay.threadDelay instance does the wrong thing for sub-microsecond timeouts.

Consider that SI.MonadDelay.threadDelay is defined over DiffTime, which has picosecond precision. But Control.Concurrent.threadDelay has only microsecond precision.

Behind the scenes, io-sim represents ThreadDelay with a DiffTime and reasons about it as if that were true. But when running the same code against IO, any sub-microsecond delay will round down to 0.

Expected behaviour

The ThreadDelay constructor should probably round any values below 1e-6 down to 0.

Desktop (please complete the following information):

• GHC version: 9.6.7
• io-sim version: 1.8.0.1
• io-classes version: 1.8.0.1

[coot]

Good point. I'd go further and just round down to the nearest microsecond.