Use better-converging MCMC kernels for default transitions

[BaxterEaves]

Currently Engine::run uses slice for row and column transitions, which can be slow to converge for large tables because they don't propose large moves. We should use both slice and sams on the rows and gibbs as default for columns.

It might also be a good idea to make set of pre-made transitions to make people's lives easier. For example, in pylace

from lace import Engine

# normal
eng = Engine.load("my-metadata.lace")
eng.run(1_000, transitions="sams-slice/gibbs")

# flat
eng = Engine.load("my-flat-metadata.lace")
eng.run(1_000, transitions="sams-slice/flat")