Allow building out-of-tree about pandocs

[mid-kid]

The third item is very quickly gonna be noticed by simply the build failing. I don't think you need to do much more with regards to that, other than either making a descriptive error, or expecting the builder to dig into the code for it (as I did, long enough to know that it wasn't gonna be a simple sed :))

[ISSOtm]

That special configuration is needed can be made obvious, sure; however, that the option exists at all may be less so! It seems likely to me that a packager will sooner reach for patch -p0 than digging through the configuration.

Then there's the issue of detection; I'd assume that ".git does not exist" should be the correct basis, but maybe I'm missing something?

Additionally, you remarked via IM that we ought to pass the --git-dir argument to git show to ensure that we don't accidentally pick up a parent Git dir. (I'm not sure if that plays nice with submodules, but my intuition says that it should.)

As for the patch, well, I think patch is indeed more appropriate than sed, yes :P
But you will have to update that patch now that #513 has landed.