remove incorrectly placed preserves_flags, and other inline asm fixes

[quaternic]

As discussed in #911

[tgross35]

Thanks! Needs to wait for #1002 but I'll merge after.

[rustbot]

This PR was rebased onto a different main commit. Here's a range-diff highlighting what actually changed.

Rebasing is a normal part of keeping PRs up to date, so no action is needed—this note is just to help reviewers.

[tgross35]

Yay for small perf improvements, a single instruction saved on memmove forward tests:

 mem_icount::memmove::forward_move large_spread_off_0:setup_forward(Cfg { len : 16, spread : Large, off ...
  Baselines:                      softfloat|softfloat (old)
  Instructions:                         299|300                  (-0.33333%) [-1.00334x]
  L1 Hits:                              420|420                  (No change)
  LL Hits:                                0|1                    (-100.000%) [---inf---]
  RAM Hits:                              12|12                   (No change)
  Total read+write:                     432|433                  (-0.23095%) [-1.00231x]
  Estimated Cycles:                     840|845                  (-0.59172%) [-1.00595x]

I think that's the test and that I actually have forward and backward swapped at

compiler-builtins/builtins-test/benches/mem_icount.rs

Lines 387 to 396 in 8ab6f08

	fn forward_move((len, spread, mut buf): (usize, usize, AlignedSlice)) {
	// Test moving from the start of the buffer toward the end
	unsafe {
	black_box(memmove(
	black_box(buf[spread..].as_mut_ptr()),
	black_box(buf.as_ptr()),
	black_box(len),
	));
	}
	}

and

compiler-builtins/builtins-test/benches/mem_icount.rs

Lines 481 to 490 in 8ab6f08

	fn backward_move((len, spread, mut buf): (usize, usize, AlignedSlice)) {
	// Test moving from the end of the buffer toward the start
	unsafe {
	black_box(memmove(
	black_box(buf.as_mut_ptr()),
	black_box(buf[spread..].as_ptr()),
	black_box(len),
	));
	}
	}

...