Ripl get_seed/set_seed are broken (and always have been)

[axch]

The correct solution is to do #138 and then expose whatever interface we want.

In the mean time:

• get_seed always returns 0
• set_seed is completely ignored in Lite, because Lite uses the ambient PRNG (until #138).
• set_seed reseeds the distinguished particle only in Puma.
• Can't just delete them right now because MRipl (#50) calls set_seed.

[axch]

Intended resolution: Delete MRipl per #50, then delete these two ripl methods, unless #138 happens first. If the latter, reconsider.

[axch]

50 being done, this is no longer blocked.

[axch]

138 is now done. Lite still ignores set_seed, but this could now reasonably be changed. Should we delete get_seed entirely, or rename it to get_initial_seed and actually implement those semantics? The initial seed is not very useful unless one also knows the computation at the beginning of which it was set, so it's not clear that it's worth storing. Another semantics we might want for a get_seed-like method would be "reseed yourself to something new and give me that seed", so that I can "save the future" relatively easily.

[riastradh-probcomp]

Once we switch to weakprng, we could change them to get_state/set_state. The vast majority of states do not correspond to any seeds -- but states are fairly small (112 bytes) and hence much easier to pass around than Mersenne twister states.

[axch]

Candidate design:

• Rename set_seed to seed_prng and make it actually work.
  • Optional: Permit omitting the seed argument, in which case it would draw one from /dev/urandom and return it. This implements the "make up a future and let me save it" functionality
• Eliminate get_seed entirely