Update wasi-libc to SDK 20

[QuLogic]

The version 17 SDK adds getpagesize, so use it instead of hardcoding a number (even if their implementation is also a hardcoded number.)

[QuLogic]

Hrm, bit confused since this test works for me locally just fine.

[aykevl]

What's the benefit here? The page size is fixed for WebAssembly at 64kB, so I don't see why we need a library call.

(This is different for getpagesize in Linux for example, as the page size varies by architecture and sometimes by kernel build - Apple M1 has 16K pages for example. There it really does need to be a library call).

[dgryski]

Aside from debating thegetpagesize change, seems like upgrading to the latest wasi-libc is a Good Thing?

[aykevl]

Yes, seems fine by me. I'm a bit careful with updating these libraries because they may introduce new bugs, but we have to do it from time to time.

[deadprogram]

@QuLogic could you please rebase this branch against the latest dev so we can verify it is passing the latest tests? I would like to merge.

[aykevl]

EDIT: oops I see I already commented on this.

---

Probably unnecessary, the WebAssembly page size is fixed at 64kB. I don't really see a benefit of calling a libc function. (This is different on an architecture like arm64 which does in fact have multiple page sizes that can only be known at runtime).
On the other hand, it probably won't increase binary size by much either so 🤷

[QuLogic]

Mostly for consistency, and to limit hard-coded numbers. I have no idea whether wasi will evolve in some direction as to not be 64KiB, but if it does, then that information will be encoded in the SDK instead of here.

[QuLogic]

Gave it a rebase. I forget which build failed, and I think logs might have gone, so we'll see if it's okay now.

[QuLogic]

I just realized there are some wasi-libc caches in CI, skipping their build if available, which is probably why CI is failing but not local.

[QuLogic]

This should be good to go now.

[deadprogram]

Thanks @QuLogic now merging.