Optimisation: S3/GS files shouldn't need to touch disk

[JensRantil]

I notice that the S3 and GS virtual file systems both copy the content first to disk. Strictly speaking, given that both APIs support range requests to fetch file content, I don't think they really need to touch disk at all. Instead they could be streamed straight over HTTP while still implementing

gleam/filesystem/vfs.go

Lines 20 to 25 in 53a6956

	type VirtualFile interface {
	io.ReaderAt
	io.ReadCloser
	io.Seeker
	Size() int64
	}

.

I guess temporarily storing to disk is a simpler implementation, though. 🤷‍♂️