Fix thread for non-unique spawns

[michael-schwarz]

The handling of threadenter in the case of a thread being spawned non-uniquely was broken:

• It failed when thread ids where non-unique.
• It actually caused a side-effect to the thread id of the creator marking that as non-unique rather than the created thread.

This PR fixes this issue which hopefully means our experiments for PLDI can continue (CC @Red-Panda64: try if merging this into your branch helps), and also increases precision for cases where the non-history thread id is enabled.

Closes #1615

[sim642]

The tests seem to be about not crashing and precision, but isn't this also a matter of soundness?

[michael-schwarz]

I thought so as well and tried to create a new test that would be unsound before but would work now. However, marking the creator as non-unique also leads to the created thread being marked as such (through the recursion in is_not_unique).
So it seems like it actually is only a precision issue vis à vis the old implementation.

Nevertheless, the original soundness tests that were added in #1187 are still around and pass.