Export interface fingerprint because whew, ABI mismatches are bad

[fasterthanlime]

Just like rubicon does essentially:

lith(22234,0x200043840) malloc: *** error for object 0xc: pointer being freed was not allocated
lith(22234,0x200043840) malloc: *** set a breakpoint in malloc_error_break to debug
error: Recipe `run` failed with exit code 134
error: Recipe `serve` failed on line 51 with exit code 134

...I just need to figure out how to.. do that fingerprinting exactly? Because... I guess something would need to run in a build script I suppose. It'd be neat if it didn't depend on the dylo cli, mhh. Need to think about this.

[fasterthanlime]

One possible way out of this is to flip the model on its head: instead of putting it onto the user to do #[cfg(feature = "impl")] for all "impl-only" things, maybe all exported things should be written as #[dylo::export] — it should be relatively simple to ignore function bodies from there, and that would be the module's signature.

the same transform would be done by dyld-cli, and load time, we'd simply check that they match. I'm not sure how to get the proc macro to "add to a registry" to get the complete signature of the module, but.. we'll get there.

maybe just exported symbols, something dirty like __dylo.{sha256sum}

[fasterthanlime]

Oooh, https://lib.rs/crates/unsynn looks interesting to do the fingerprinting maybe