Resolve remaining differences between mainline elfutils and our fork

We should reconsider the remaining differences between mainline elfutils and our fork and try to get rid of them. In particular:

• Link the backends into libdw and get rid of the dlopen() call and the need to do rpath. This allows us to drop the incredibly evil windows specific hack that ties perfparser to a specific working directory. Also, it would be acceptable upstream.
• Resubmit the --strip-trailing-cr patch. It seems to have fallen through the cracks.
• Resubmit the err/errx patch. It wasn't applied but also not rejected
• Come up with a better solution to the O_BINARY problem. This is a maintenance nightmare.
• Do something about the gnulib stuff. Can we simplify this somehow? Can it be an external dll? Can we maybe make use of the linux-userland-on-windows-kernel stuff that's surfaced recently? Then we could use an actual glibc and be done with it.
• Resubmit the stack size patch