Trustless mining

[dzwdz]

Solves #60.
It's not really mergeable yet. The current issues are:

• I only tested this on Alpine Linux, and didn't exhaustively check all configurations for possible regressions etc.
• I had to disable ref10 support, because it lacked some essential functions iirc. Hopefully this won't be too big of an issue, as ed25519-donna should be a complete replacement? Maybe? Are there systems where it doesn't run?
• I had to pull in ed25519_impl_{pre,post}.h in main.c, which is pretty horrible.
• Nobody else looked at the crypto yet but me - I think the math checks out, but you shouldn't rely just on me.

I tried to make it somewhat idiotproof by ensuring you won't get a private key that'll work with Tor unless you did everything right. I did not protect against someone using the same base key for multiple .onions - maybe I could add a disclaimer when using --combine if you think that's necessary.

Thanks in advance for any feedback ^^

[cathugger]

regarding ref10 not having proper functionality, "correct" solution would be enabling trustless code only when configure flag is set to true, which could be true by default & conflict with ref10.
or removing ref10 as I'm honestly not sure what it's ever good for.
it was first method supported and is still possibly more portable than donna as the code at least looks less ifdef-y and cleaner to me, but whether that is worth much in reality is different question.

whatever you've done here right now is probably fine for now, I can make decision later.

[cathugger]

I'll dig into this deeper when i have more time.