Make leb64 encoding fully safe

[jonas-schievink]

This removes all unchecked indexing as well as the usage of MaybeUninit::uninit().assume_init() from the leb64 encoding logic.

Resulting binary size change for the log binary:

   text	   data	    bss	    dec	    hex	filename
  33708	      0	      8	  33716	   83b4	log-main
  33508	      0	      8	  33516	   82ec	../../target/thumbv7m-none-eabi/release/log

That's right, it's 200 Bytes smaller.

I've also tried keeping MaybeUninit in place, which reduces binary size by like 400 more bytes. That seemed spooky, so I didn't go for that (zeroing 10 bytes shouldn't take 400 Bytes of code space? does it pull in memset? does it miscompile something due to the UB?). Haven't investigated that further.

[japaric]

This is better. Thanks.

That's right, it's 200 Bytes smaller.

was that with opt-level=3? maybe the indexing prevents loop unrolling and that makes the binary smaller? I would have no idea why keeping the assume-uninit makes the program that much smaller though. llvm optimizations are unpredictable IMO

[japaric]

bors r+

[bors]

Merge conflict.

[jonas-schievink]

was that with opt-level=3?

It was cargo build --release in the firmware/qemu directory, which should be opt-level=3, LTO=fat.

[japaric]

needs a rebase but feel free to submit to bors once that's resolved

[jonas-schievink]

bors r=japaric

[bors]

Build succeeded:

• ci