Use mpint representation for shared_secret when deriving keys in pure-PQ key exchange, and some other bug fixes; fixes #119

[kevinmkane]

Other bug fixes:

1. Increase MAX_PROP to 160. Large numbers of proposals due to all the OQS algorithms can cause a non-spec-compliant selection of algorithm if this value is too low and client and server order their proposals differently. This had been fixed in v7 by Increase MAX_PROP to 140 #85 but returned in the v8 branch.
2. When calling ssh-keygen with the -t parameter to specify the key type for one of the ECDSA hybrids, but using the short name instead of the full ssh-ecdsa-xxxxxxx name, ECDSA keys would always be generated on P-256. Instead, always use the specified curve.
3. kex_kem_generic_{enc,dec} could use an uninitialized variable as the return value if the call to the underlying liboqs function failed. On the small chance that variable happened to have a zero in it, this would look like a successful result to teh caller. Return SSH_ERR_LIBCRYPTO_ERROR in these cases instead.

[kevinmkane]

3. kex_kem_generic_{enc,dec} could use an uninitialized variable as the return value if the call to the underlying liboqs function failed. On the small chance that variable happened to have a zero in it, this would look like a successful result to teh caller. Return SSH_ERR_LIBCRYPTO_ERROR in these cases instead.

Actually I'm wrong about this: it's worse. The r variable will be initialized, but to zero based on previous calls succeeding that assign to it. So it's certain failures in the calls into liboqs won't get caught. The fix is still correct, fortunately.

[dstebila]

Thanks @kevinmkane!