Smooshables need likelihood scalers about linearham HOT 5 CLOSED

Renaming to "scaler" to correspond with libpll.

from linearham.

• scalers are exponents
• to get to the correct log likelihood, we need to subtract by some multiple of the scaler (https://github.com/xflouris/libpll/blob/dev/src/core_likelihood.c#L353)

Also, no need to have scalers in germline.

from linearham.

Given that we may be looking at zero likelihoods from dummy paths, it seems to me that we should have an assertion in the rescaling code that keeps us from getting an infinite loop in the case of exactly zero likelihood.

from linearham.

Since the 0's will be within matrices that have non-zero values, I don't see how the infinite loop happens. Right now, the SCALE_FACTOR only multiplies if all entries of the matrix are less than SCALE_THRESHOLD, and since our matrices are mixed, we don't ever have a matrix just with 0's so it will stop at some point?

from linearham.

Would still be good to have an assertion that matrices aren't all zero.

from linearham.