SharedArrayBuffer atomic input?

[ChALkeR]

Currently, we accept Uint8Array<SharedArrayBuffer>, but operate on that assuming there are no races
When handled byte-by-byte, the input might get replaced from another thread

We should likely copy in an atomic way, even if that hurts perf - that's what whomever is using SharedArrayBuffer is likely expecting anyway and is the safer behavior

See also #66

[ChALkeR]

From checking the spec and writing a quick test, even built-in operations on SAB (including SharedArrayBuffer.slice()) are non-atomic.

It's the user responsibility to install locks / ensure no races there, so we are good just handling SharedArrayBuffer input as regular ArrayBuffer.