61e37ce2

By: ValentinHirschi <valentin.hirschi@gmail.com>

document the implemented h-shaped inverse CDF

Describe the normalized log-logistic mixture, its bracketed solve in log(t),
and the Jacobian obtained from the same proposal derivative. Keep physical h
and h_dual unchanged and record the user's decision to retain this construction.

Validation: checked the documentation against the sampling map and eager dual
implementation; git diff --check passes. Runtime optimization and its saved-state
acceptance work remain separate uncommitted changes.

User request:
How do you do the inverse CDF for the cutkosky cut h-function. It does not have a closed analytical expression.
Also keep in mind that it is not absolutely necessary to have the exact h-function inverse. It is also ok to have a simpler function that mimics this shape close enough. But anyway it seems like it's not longer the bottleneck, so you can keep what you have already.

Timeline