Consider thread-local variables non-unique

[michael-schwarz]

Fixes #1072 and the unsoundness described there.

[jerhard]

Does marking thread-locals variables as non-unique also lead to them being only weakly updated? If so, could one treat thread-local variables as non-unique only when they are escaped, to remedy this issue?

It should not be possible to have address comparisons between multiple instances of a particular type of thread-local variable without one of them having escaped.

[michael-schwarz]

Does marking thread-locals variables as non-unique also lead to them being only weakly updated? If so, could one treat thread-local variables as non-unique only when they are escaped, to remedy this issue?

The weak update mechanism applies mainly to locals, here the variables are still treated as globals (pending a more systematic approach). However, I added a test whether this thread-local variable has its address taken, which was also the proxy we used for having escaped in #1071. I tried for a bit, but could not really construct any examples where that really makes a difference. I still think it is good to have it in the interest of consistency though.