Also update the docs to point at those instead of our thin/EM patches.
They actually count from 0, not 1. I hadn't originally intended for this to be the behavior, but now that the code is written, it seems preferable to attempting to do a bunch of conversions between 0-indexing and 1-indexing, so I think I'd prefer to keep it.
The space-separated EINHORN_FDS interface was an attempt to pretend to be compatible with Upstart. Upstart internally supports passing multiple FDs, but exposes no interfaces that would allow it. Meanwhile, having the space-separated list of FDs makes some other applications more difficult.