When the augment() documentation was first written, evidently matrix
columns were not allowed in tibbles. But they have been supported since
tibble version 1.0.0: tidyverse/dplyr#3630

For instance, augment.lm() works correctly for a model containing poly()
or ns() terms, placing those terms in matrix columns as expected.

This was forbidden in broom 0.5.0 because tibbles did not support matrix
columns, and Surv() produces matrix columns in the model.frame(). But
since tibble 1.0.0, matrix columns are supported, and these work fine
with the default model.frame().