encode backward and use C stubs

[c-cube]

rationale: encoding backward means we can use a single buffer even in the presence of nested messages. We encode from the end, in reverse order, which means we encode the sub-message before having to write its size; so when it's time to write the size we know it and we can just write the varint. If we write forward we cannot do that because we don't know how much space to leave in front to encode the length as a varint.

The C stubs are there to make varint operations faster: computing how many bytes a varint takes (so we can allocate a little slice for it in the buffer); and writing the varint into this slice.

Note that we also require OCaml >= 4.08 so we can rely on Bytes' operations for fixed-size bits32/bits64.