Fix SourceFile::normalized_byte_pos

[eggyal]

This method was broken by 258ace6, which changed self.normalized_pos to use relative offsets however this method continued to compare against an absolute offset.

Also adds a regression test for the issue that this method was originally introduced to fix.

Closes #149568
Fixes regression of #110885

r? cjgillot (as author of the breaking commit)

[eggyal]

@rustbot author

[rustbot]

Reminder, once the PR becomes ready for a review, use @rustbot ready.

[eggyal]

The test was failing on aarch64 because it uses a different asm syntax/language, so my # prefixed comments were rejected. This has been fixed by replacing with // prefixed comments, which are valid in both the Intel and Arm asm syntaxes.

However, differences in the precise error output required manual normalization. I suspect we should instead enable this test just for Intel architectures? Otherwise there will no doubt be other syntax/normalization failures on other targets?

@rustbot ready

[rustbot]

Some changes occurred in src/tools/compiletest

cc @jieyouxu

[eggyal]

Okay, so it was then failing on gcc which again isn't a surprise. So I've reverted the test normalization and restricted it to x86_64 on llvm; however the needs-backends directive (introduced in #144125) wasn't being recognised—hence 7e39015 (which should probably be in a separate PR? but it's soooo minor....).