Fixed htmlReadFd() + added regression tests for htmlReadFd() and xmlReadFd()

htmlReadFd() has been using a mixture of xml and html context handling functions, meaning it doesn't work properly and fails to parse most html. This seems to have gone unnoticed, since I can find no record of anyone actually using this function.

I have also added regression tests for htmlReadFd() and xmlReadFd(). Had these been in place beforehand they would have caught this error.