investr 1.5.0

• New invest() methods for "survfit" and "Surv" objects (from the survival package) estimate the survival time at which a Kaplan-Meier curve reaches a given survival probability (e.g., median survival time), with confidence limits obtained by inverting the curve's pointwise confidence band (#30).

• invest() results with interval = "percentile" now also inherit from class "boot", so they work directly with functions from the boot package such as boot::boot.ci() (#32).

• nlme moved from Imports to Suggests; "lme"-specific code paths now check for it with requireNamespace() (#47).

• Fixed calibrate.lm() silently mishandling models with poly()-transformed predictors; it now reads the predictor from the model frame and errors informatively instead (#48).

• Fixed the formula for g in the inversion-interval vignette, which had been broken by a regression in a prior "typo fix" (#51).

• Fixed predFit() with se.fit = TRUE failing on "nls" models with more than one predictor variable (#53).

• Fixed invest() and predFit() failing when the fitted model was created inside a function, or with no data argument at all; training/prediction data is now reconstructed via model.frame() or the environment captured by the model's formula, instead of evaluating object$call$data in the caller's frame (#41, #42, #45).

• invest() now validates that newdata columns have classes (and, for factors, levels) compatible with the fitted model's training data, rather than failing with a confusing downstream error or silently returning a nonsensical result (#36).

• The "Use plotFit for guidance" hint on a failed confidence-limit search no longer appears for multi-predictor lm/glm models, since plotFit() doesn't support more than one predictor variable (#35).