Like you point out in the future directions, I too worry about the feasibility of publishing the macro binaries for package authors. IIUC, it puts developers with Linux hosts at a disadvantage since macOS isn't supported as a target platform unless you're on a macOS host? And I wouldn't know where to begin as a macOS user if I wanted to support Windows hosts.
There was a proof-of-concept a couple years ago that built macros for WebAssembly. Is that something we should be investing more in? Anyone can cross-compile to it, it would mean that package authors only have to target a single host for prebuilts instead of many, and we'd get some other things like sandboxing for free.
The implementation changes you linked to are small enough that I think it would be silly to say we shouldn't add the support you're proposing, but I wonder how much it realistically moves the needle. Packages that target iOS/macOS/watchOS/tvOS/visionOS probably benefit the most from this since building a macOS binary is trivial for those developers, and maybe that's a large enough group to justify it.
I suppose the code signing requirements for these executables are the same as what's already required for .xcframeworks?