Point::steg_encode was leaving the 24 high bits of the buffer as zero.
It also ignored the size parameter.  The size parameter has now been removed,
the zeros fixed and a test added to make sure that it is fixed.

Per https://github.com/MystenLabs/ed25519-unsafe-libs, deprecate eddsa signing
with separate pubkey and privkey input.  Instead decaf_ed*_keypair_sign

Bump version number in CMakeLists.txt, thanks Johan Pascal.
Change test scripts to avoid GCC warnings, thanks Sofi Celi.
Add full RFC 8032 test vectors, thanks Sofi Celi.

* SunOS linker doesn't support --gc-sections
 * Add portable_endian.h with __sun version of htole64 and le64toh
   * Replace portable endian code in shake.c with inclusion of portable_endian.h
   * Replace portable endian code in word.h with inclusion of portable_endian.h
 * Add explicit extern reference to word.h for posix_memalign when __sun defined
 * Replace references to u_int*_t with uint*_t
 * rdtsc call in shake.c was only working on 32-bit i386
 * rdtsc call in bench_decaf.cxx was inaccurate on 64-bit x86_64 when clang absent
 * Fix two signed/unsigned comparison errors in test_decaf.cxx

Before, invert_elligator would invert to a gf, which wouldnt be a uniformly
random string because, eg, curve25519 gfs only have 255 bits out of 256.
Now add a random multiple of p.  This still wont work for future curves
that have a field size of 1 mod 8, because those curves use elligator with
no high bit set, but its a start