Fix beta vs. master differences

[Mark-Simulacrum]

Despite having added #11, we still saw during the 1.54 beta promotion that master behavior on a particular test seemingly differed from that observed on the beta channel. In particular, rust-lang/rust#86413 was only able to merge with the pgo-branch-weights test disabled, even though this test worked on master at the same time.

We've also been applying patches to workaround non-working tooling (e.g., RLS) on aarch64 MSVC -- rust-lang/rust@c4aeccd only on beta.

Ideally the beta promotion should be much smoother -- without need to backport patches to make it work. We should figure out how to help make this happen, perhaps by adding another -beta builder like we did to fix #11, though this feels a little wasteful.

[pietroalbini]

I think we should definitely fix the aarch64 MSVC RLS issue on master too by disabling it.

I'm not sure whether a -beta builder would've actually made a difference in catching the PGO bug: the bug was happening on aarch64, which we aren't testing on stable either.

[Mark-Simulacrum]

Oh, I see your point. Perhaps adding a stable builder for aarch64 makes some sense (as the primary other tier 1 arch).

[Mark-Simulacrum]

But more broadly, working to make stable/beta builds closer to those on nightly/master seems like a good idea. It's concerning that a test like pgo branch weights even is affected - for unknown at this time reasons - by the beta branch.

[pietroalbini]

I'm really interested to know why that bug showed up just on beta, and I'm fully on board with adding something to our CI catching similar issues in the future. I wouldn't pre-emptively add other -stable builders for other OSes/archs though.