Fix some cases of UB

https://rust-lang.zulipchat.com/#narrow/stream/136281-t-opsem/topic/How.20does.20the.20GCC.20backend.20handle.20code.20C.20would.20disallow.3F

We probably have:

• clobbering neighboring padding of assigned field

(jakub:

- AFAIK even C++ ought to guarantee state of padding bits in zero-initialization (but undefined when copied over e.g. through structure copy rather than memcpy; but our CONSTRUCTOR right now doesn't have a flag which would say all the padding is zero initialized
- and the question is what SRA etc. will do for it
- if all the Rust struct initializations are supposed to e.g. zero all padding bits, rather than say just small subset of them, then either the middle-end should be adjusted e.g. to have a flag on CONSTRUCTOR which guarantees it, or indeed create fields for all paddings and zero initialize them (all the FEs use stor-layout.cc routines for laying out structures/unions, so one can use that
- and then see where the paddings are and deal with those
- or there is __builtin_clear_padding builtin...

)

• pointer lifetime end zapping
• partially uninitialized data in union has to be preserved when it's moved around.

Maybe those will help find out what's currently UB:

• -fsanitize
• -fanalyzer: https://gcc.gnu.org/onlinedocs/gcc/Static-Analyzer-Options.html

Start with the linked Zulip discussion and investigate the listed cases: padding clobbering, pointer-lifetime end zapping, and preservation of partially uninitialized union data. Use -fsanitize and -fanalyzer to identify current undefined behavior; done requires narrowing the cases to reproducible findings and addressing the confirmed issues.