release 0.7.0

[hannesm]

issues to address:

• run-time instead of build-time feature detection for aesni (issue instead of compile-time cpu feature detection (of AESNI etc.), do load-time (run-time) feature detection #45) done via Runtime detection of CPU features #53
• revise entropy / PRNG
  • unless there are convincing users of fork() and our prng, revert to use fortuna (seeded by getrandom/getentropy) on unix see RNG and entropy improvements #64
  • revise periodic entropy feeding (re-enable on unix with lwt; revise to not use rdrand/rdseed at every iteration - maybe "at most every 100 ms" or something similar) -- see https://www.cryptopp.com/wiki/RDRAND_and_RDSEED who measured the cycles of rdrand/rdseed (see RNG and entropy improvements #64)
  • adapt fortuna accumulation / feeding to its design document (esp. Accumulator.add always sets the trap (should be guarded with minimal pool size, and an upper bound in respect to timing (once each 100ms))) (see RNG and entropy improvements #64)
  • find and/or adapt test vectors for fortuna and entropy feeding (see https://www.seehuhn.de/pages/fortuna#sec:4.1.0 and https://github.com/dlitz/pycrypto/blob/master/lib/Crypto/SelfTest/Random/Fortuna/test_FortunaGenerator.py)
• integrate Variants #48 for cross-building for mirageos4 (on hold)
• point from "unseeded generator" exception to how to seed the generator (either in docs or the exception itself) -- happened e.g. on mirage if you use Mirage_crypto_rng.generate without using the random device in your unikernel (ssh-agent had this issue)
• make powm_sec optional (i.e. being able to disable it), there are scenarios where paying the 20% isn't worth it -- esp. since it does not lead to "all is constant time now"
• replace ct_find_uint8 by eqaf (>= 0.7.0) done via Replace ct_find_uint8 by Eqaf_cstruct.find_uint8 #52
• 32bit support (requested in Compile errors on aarch32 #60) -- done in fix 32 bit support (ghash_ctmul: a ghash implementation from BearSSL that works on 32 bit LE platforms) #65

feel free to comment on this issue about more tasks to include in the next (major) release.

[hannesm]

most issues are adressed, I'll cut a release (the remaining are test vectors for fortuna, which can be addressed separately, and only optionally use powm_sec, which can as well wait until requested)