Add garbage collector that uses an external allocator

[niaow]

In cases where tinygo is running in an environment that it does not own, it may be desirable to allow tinygo to use a C heap. This is accomplished by prepending treap nodes to the beginning of every allocation.

Marking a node is implemented by removing it from the treap and then pushing it onto a scan queue (which is implemented as a linked list using the memory previously used for the treap node). Nodes are then popped off of this queue, scanned, and then pushed onto a new treap. Once the queue is empty, all currently active nodes will have been moved from the old treap to the new treap. All nodes from the old treap are then destroyed, and the new treap is swapped in.

All operations are done non-recursively. This is still somewhat WIP, and there is a lot of debugging code still left in right now.

[niaow]

This PR should be ready now.

[aykevl]

Looks good to me, although I can't say I fully understand it (treaps are foreign to me).

One issue though: this needs some sort of testing. Ideally ./testdata/gc.go on the host, but those integration tests are currently a bit inflexible. Until then I think it would be best to include a smoke test with this GC on the host OS to see whether it compiles at all.

[aykevl]

I only now spot this freebsd,!baremetal in the code, which doesn't make a lot of sense. The fact that there is linux,!baremetal is because baremetal targets pretend to be linux/arm to get the standard library to compile. Such a !baremetal exclusion is not necessary for FreeBSD.
But this is not something that needs fixing in this PR.

[niaow]

Ideally ./testdata/gc.go on the host, but those integration tests are currently a bit inflexible.

This has actually turned out to be surprisingly effective, and was able to catch tons of bugs.

Until then I think it would be best to include a smoke test with this GC on the host OS to see whether it compiles at all.

I am not quite sure what you mean. Don't the existing tests already check that they compile?

[aykevl]

Until then I think it would be best to include a smoke test with this GC on the host OS to see whether it compiles at all.

I am not quite sure what you mean. Don't the existing tests already check that they compile?

Ah, now I see. You also switched the default GC on Linux etc. Then it's fine.

One thing I only now notice: some documentation doesn't include the new GC type, such as the help text in the tinygo command (main.go) and the function documentation for (*compileopts.Config).GC.

[niaow]

Alright, I think I updated all of the docs in this repo.

[aykevl]

Thanks for updating!