Add support for UNIX domain sockets

[kentslaney]

interface for memcached -s

[kentslaney]

I think this is ready to be squash and merged. I can rebase the branch and force push myself if that's easier.

The only existing interface I modified was connection::connectPoll which is protected anyway.

The other notable change is that if you're using make test instead of the /misc/travis scripts, test_unix.cpp expects that memcached was started using ./misc/memcached startall instead of ./misc/memcached start.

The PR workflow is still awaiting approval, but in the meantime the forked workflow results can be compared to the origin/HEAD workflow baseline. I still haven't gotten act running properly which is why there were so many commits towards the end.

[lexdene]

I have fixed ci on master branch.

Would you please merge (or rebase) master first to see if unittest can pass?

[pypeng]

need a version bump, like this: #116

[kentslaney]

The server string implementation got more complicated than I expected and there's still dead code in there. It should get optimized out by the compiler. I don't think it'll present notable maintenance overhead, but would also understand a refactor request.

[kentslaney]

After looking more closely at the default branch I realized there was a reason it was parsed the way it was. The PR implementation is still more complicated than the initial binding language solution but seemed worth it for the guaranteed language consistency. There is no longer dead code.

[kentslaney]

One more bump on workflow approval please @lexdene (I assume? Can't tell who approved it). Looks like this was the issue. Works on my fork this time, should have caught the last one.