cfail tests do not check for unexpected errors on windows

[Diggsey]

Currently the cfail test runner works by parsing rustc's stderr, and scanning the output for errors/warnings. It uses the pattern <filename>:<line>:<col> [error|warning] to check whether a line of output is an unexpected error.

This never succeeds on windows, because absolute paths start with a drive letter + colon, eg. "C:", and the colon breaks the pattern.

The result is that any number of additional errors are allowed to be generated without causing the cfail tests to fail.