memo/planbuilder: Off by one error, and issue importing statistics for testing

[max-hoffman]

The original bug was a one-off error where conjCollector was checking for the wrong equivalence column. That spiraled into:

• statistics were not properly being used for some tests because PRIMARY case sensitivity in ANALYZE ... UPDATE HISTOGRAM statements
• one integration test had a weird edge case where OSX and linux compute slightly differently float values, leading to different plans locally vs CI. One option was to truncate the specificity of all cost values with a multiply/round to avoid numerical weirdness...picking the first lowest plan also seems to work.
• variety of other small bugs in costing
• update join bug where the projections after a rearrangement lost expression id/nullability info. This was exposed by the changes above