refactor analysis into a single ccn2019 notebook.

fix VL bug, and use lag to shift local variables so for every trial we only consider previous local params.

remove current trial from local stat calculations.

init a new dummy var analysis for ccn2019

refactor modeling "correct" to another script