Fixed a pkcs#1 v1.5 flaw regarding leading zero bytes

With these changes the entire new pkcs#1 test suite passes fine.

The leading zero bytes used to appear due to mixed signed/unsigned use
of our internal representation of the values as MPIs. The changed code
also detected another bug in the DSA selftest which used the pkcs1
flag - this was certainly wrong but didn't throw an error. The code
in GnuPG does the right thing thus I believe not too many applications
got it as wrong as we in our own selftest.