Address CTFE perf regression introduced by #71005

[jonas-schievink]

In an effort to treat the return place _0 more closely to regular locals, #71005 has removed some optimizations that avoid allocating memory for locals when there's only a return place. This regresses the CTFE stress test by around 8%.

It would be good to gain the lost performance back. #71005 (comment) contains some ideas for how to do that.

[RalfJung]

Regarding what I wrote there:

The main subtlety here is making sure this is a Miri-internal optimization that cannot be observed by the program. The fact that we reborrow the return place should exclude any observations by accessing the return place directly... so I think the only remaining possible observation is through address equality tests.

I think address equality is actually not an issue as the program cannot observe the address of the return place. There is no way to write &_0 in surface Rust. (Optimizations can introduce this but of course not in ways that are observable.)

So, yeah, that plan should be fine actually.