[apple] support watchos_arm64 in toolchain

[dflems]

The watch-only changes from #14439

[dflems]

not sure if this will conflict with the one below or if there's a different CPU that should be used for one or the other?

[aiuto]

It looks like it may in the future. If the constraints are the same, toolchain resolution would pick this one first.
But, what is up with the name watchos_arm64_32?
That entry is strange with a name that implies 32 bits, while using aarch64.

[brentleyjones]

arm64_32 is a variant of arm64 with 32-bit pointer sizes, used on Apple Watch Series 4 and later.

[dflems]

I wasn't sure about these. But ios_sim_arm64 and ios_arm64 are identical here too, so if it's wrong what's in here already is wrong.

[dflems]

cc @Wyverald @kaylathar I broke this out of that other PR (#14439). The watchos_arm64 CPU already existed. This PR doesn't add any new ones, just integration with the default osx toolchain.

[kaylathar]

This LGTM - no concerns

[brentleyjones]

@oquenchil or @Wyverald can you import?

[aiuto]

Not really a review. I don't have enough knowledge of the Apple naming conventions.

[aiuto]

It looks like it may in the future. If the constraints are the same, toolchain resolution would pick this one first.
But, what is up with the name watchos_arm64_32?
That entry is strange with a name that implies 32 bits, while using aarch64.

[Wyverald]

I'll do the import.

[Wyverald]

@katre asks:

Not @platforms//os:watchos?

I don't know why they are all set like this, but this feels like a misconfiguration to me.

[brentleyjones]

I don't think this change should change those, as that's a much larger and riskier change for this current addition.

[katre]

Someone should investigate these, but I don't know enough about the difference between the ios variants to know what's correct.