Fix unittests for windows 32bit intel

[matthiasblaesing]

With the improvements done by @tresf in #1264 three failing windows unittests on x86 were identified:

• GetModuleFileNameExW can fail on 64bit platforms, if the calling programm is 32bit. I did not find official documentation about this, but the indication I found on the net speculate, that the problem is observed, when the target process is mapped beyon the memory range, that a 32 bit process can observe. GetProcessImageFileName can be used as an alternative, but returns a different format (device path) and needs to be converted, to return the expected value.
• testCreateRemoteThread creates a thread by writing the binary directly into memory and starting the thread with the memory location. That binary was only valid for 64bit, but nor for 32 bit (different calling convention). The binary for x86 was retrieved by combining observations of binaries and assembler output from mingw and reading the opcode documentation for x86.
• testWindowMesssages failed because SetWindowLongPtr only exists on x86-64, on x86 it is only present as a header based alias for SetWindowLong

[matthiasblaesing]

@dbwiddis could you please have a look?

[dbwiddis]

Left some comments on the first branch. The second two look sane to me but I'm not experienced enough with inline assembly or callbacks to conduct a more substantive review.

[matthiasblaesing]

Thank you for the review Daniel. I admit, that I did not want to do that many changes, but you are right, using the intended function is the sane approach. It might also help to fix #1269. Could you have another look, I had some WTF moments looking at the various path options windows offers.

I'll take care of the merge conflict once this is merge ready.

[dbwiddis]

Looks good to me!