Hide mutex structure contents and standard globals in base analysis

[sim642]

Split off from #745 (comment).

@michael-schwarz:

I'm a bit worried about having Bot here, as it might make some comparisons of the struct members return false, potentially making both branches unreachable.

@sim642:

Maybe a safer alternative would be like a Struct with no fields? Of course introducing a new Mutex variant to the value domain would be the safest, but I suspect it requires similar hacks as for Thread since there's some unknown structure still being manipulated (and that structure probably is different for OSX).

---

This is to avoid large amounts of invariants about pthread_mutex_t contents, which are actually all incorrect, since base analysis never updates them from their initialized values. Hiding them from elsewhere (e.g. g2html) would improve readability as well, but so far such mutex type ignoring was only done for accesses, not base.

Additionally, it (optionally) hides unknown globals from standard headers, e.g. __tzname, etc, such that they wouldn't always clutter local states.

[michael-schwarz]

LGTM! We might want to enhance this mutex data type with information such as whether it is initialized or whether it is recursive at some later point, but this already is a step in the right direction!