-------------------
- Major change to HMM fitting
- Now convergence is checked by looking at the max log likelihood over
iterations and determining when the gaussian filtered trace has plateaued
- Also implemented HMM rejection if fitting results in decreasing log
likelihood when fit with linear regression after 100 iterations
- Further HmmHandler, after fitting and hmm, will roll back the HMM to a
local maxima in log likelihood within a plateaued region
- Added useful function to poissonHMM to check trend of log likelihood trace
and help determine if an hmm has plateaued