Well, a few days ago there was a post about the Unison language, where code is stored in AST form and text is just a representation.
You could then decompile to some alternative syntax, but you'd lose any idiosyncratic formatting represented by the compressed diff.
Last I checked, Java AoT compilation precluded runtime re-optimization, though I presume they've fixed that by now.
Last I checked, they both used stack-based bytecode, which typically takes longer to JIT and results in slower native code than a compressed SSA / control flow graph (see the SafeTSA papers).
Though SSA is deferred to JIT/ILC instead. In either case you get the access to all the actual low-level bits when you need to. No other portable target lets you do that.
I don't see why it couldn't be done though, I think it just hasn't been a priority. Heck, you could have 100 different users collaborating in 100 different "languages", and so long as they serialized to the same AST and back, none of them would ever have to see the atrocious syntax which the other users prefer. Their editors and browsers could just render everything according to their users' preferences.
Edit: it appears that Unison has an issue for this feature: https://github.com/unisonweb/unison/issues/499