fix(compiler): Write universal exports on linked module

[ospencer]

In #2104, the write_universal_exports function was previously called after linking, but I forgot to reinstate it. This restores that behavior.

I deleted some tests because they were testing old behavior that was valid for the intermediate (object) wasm files but are no longer valid in the output of final wasm modules.

[spotandjake]

Mostly looks good just two small questions.

[spotandjake]

I was testing with this locally and I think we changed the behaviour of the start function.
The below program:

module Test

provide let _start = () => print("test")

when compiled with grain compile ./test.gr --use-start-section gives an error during compilation Fatal: Module::addFunction: _start already exists, oh actually interestingly I get that even if I omit the --use-start-section

We also get similar behaviour with provide let memory = () => 1 because of the memory section export.

[ospencer]

@spotandjake Good catch. I restored the old behavior. We should maybe consider a warning for reserved export names, but that's not for this PR.

[spotandjake]

I just tried this again the _start stuff is fixed but I found a new bug if you have a program like:
b.gr:

module B

provide let x = (y) => 1

a.gr

module A
from "./b.gr" include B
provide let x = B.x

we seem to hit a Not_found exception.

It might also make sense to add regression tests for both the _start and example above.

I did some independent investigation on this and the problem is B.x is being imported as a global, and is in the export table as a global.

[ospencer]

This is good to go. @spotandjake I added tests for that behavior and another test caught a different bug.

[alex-snezhko]

Why is the remove_export no longer necessary

[ospencer]

When we used wasm files as object files, we wrote special Grain export names in the Wasm binaries to facilitate linking later. We don't do this anymore before linking happens before we generate wasm (so there was no Grain export in the first place to worry about removing).

[alex-snezhko]

Why is the GRAIN$EXPORT$ prefix gone?

[ospencer]

We don't write special Grain exports anymore; now we only write the universal exports (see my other comment).

[spotandjake]

Looks good to me I had one small question, but otherwise lets get this merged.

[spotandjake]

Should we be doing a check here and warn and override if a user _start exists?

[ospencer]

Maybe, but that's not for this PR.