feat(c-api) Add wat2wasm to the wasm-c-api module

[Hywan]

This PR adds the wat2wasm function to the wasm-c-api module.

Caveat of this patch: The wasmer_wasm.h seems to be written by hand. Consequently, if the lib/c-api/ crate is compiled without the wat feature (which is the default), the wat2wasm function will still be defined in wasmer_wasm.h.
Proposal: We must generate the wasmer_wasm.h header file with the same approach than we use for wasm.h. It can be done in a next PR.

Thoughts @MarkMcCaskey

[MarkMcCaskey]

We discussed organization of the C APIs privately, and I think with some of the changes we talked about we can probably generate that file, but for now it's non-trivial to do so.

I'm unsure if we want to make this transformation explicit or not. In Wasmer we allow taking a &[u8] and we'll handle both wat and wasm. Perhaps we just want to do that for the C API too.

If we want to make it explicit, then this change seems fine. How would users free this data though? And one more point, when turning things like Vec or String into a pointer + length, it's important to shrink_to_fit first so we don't leak memory. Vec has 3 numbers: pointer, length, and capacity. If capacity != length (which is very likely), then we can't reconstruct the Vec correctly and will probably do a partial free

[Hywan]

I've rebased this PR on top of the master branch.
I've also added shrint_to_fit.
Finally, I've changed the wat2wasm function type signature:

- unsafe extern "C" fn wat2wasm(wat: &wasm_byte_vec_t, wasm: &mut wasm_byte_vec_t) -> Option<()>
+ unsafe extern "C" fn wat2wasm(wat: &wasm_byte_vec_t) -> Option<Box<wasm_byte_vec_t>>

I believe this API is better. If it returns NULL, it's clear there is an error; Also it helps to use ? in the implementation. It also makes explicit that the callee is responsible of freeing the data.

Thoughts?

[MarkMcCaskey]

Looks good! Other than some small bugs it looks good to go! One other thing, perhaps we want to add an example/test for this function (it may have caught the use after free bug, though it's highly likely it would not have)

[Hywan]

bors r+

[MarkMcCaskey]

Looks great!

[bors]

Build succeeded:

• Audit
• Code lint
• Test on linux
• Test on linux-aarch64
• Test on macos
• Test on windows