The `open()` commands in `makedep` for reading Fortran source now
includes an `errors=` argument for catching bytes outside of the file
character set.  Unknown characters are replaced with the "unknown"
character (usually �) rather than raising an error.

This avoids problems with Unicode characters and older Pythons which do not
support them, as well as characters from legacy encodings which can
cause errors in Unicode.

Substitution does not break any behavior, since Unicode is only
permitted inside of comment blocks and strings.

This fixes several errors which were silent in `.testing` but were
observed by some users which using autoconf to build their own
executables.