computeForestLeafIndices() doesn't check for propensity covariates in bcf models

[jaredsmurray]

computeForestLeafIndices() takes a covariate matrix, but not a propensity score and doesn't check model_params to see if the prop score was used as a covariate like predict.bcfmodel does. The return value has the expected dimensions and plausible content but is off in small, upredictable ways when the propensity score was used as a covariate (presumably because the function walking the tree is going out of the bounds of the supplied covariate matrix when the tree splits on the prop score). Manually appending the propensity score or omitting it as a covariate corrects the issue. The fix is probably to borrow the logic and preprocessing in predict.bcfmodel.

[andrewherren]

This should be fixed now! You can pass a propensity argument to the computeForestLeafIndices() function (and if the BCF model has an internal propensity model, it will predict from that instead, mirroring the predict.bcfmodel logic as you noted)