[patch/idea] Add register scrambling to testsuite

[Florian Weimer]

* Wilco Dijkstra via Libc-alpha:

> Also this is the first time I've heard about an uninitialized read - the
> majority of bugs in string functions are related to page cross handling,
> reading before/after the input, overflow errors of the 'n' parameter, basic
> logic errors or selecting wrong ifunc. Why should we do something about
> uninitialized variables when the testsuite could be improved to better test
> all these corner cases?

It seems to have a favorable results/effort ratio.  Perturbing the
registers during test startup (instead of passing through the zeros that
come from the kernel) shouldn't be that difficult.  Crucially, it also
does not drive up the subvariants to test, increasing testing time.

Thanks,
Florian