Nice get_draws(shape = "DxP") column names break when response is multivariate

[andymilne]

(See #1318.) For ordinal models in brms, posterior_predict (called by type = "prediction" in marginaleffects) averages over the levels of the response variable, but posterior_epredict (called by type = "response" in marginaleffects) does not. In the latter, we get a probability for each response level, which are called Group in marginaleffects. In the below code, marginaleffects generates nice column names for the former but not the latter. For the latter, I suggest concatenating the Group number to the remaining factor levels to make nice column names.

data <- mtcars

mdl <- 
  brm(
    carb ~ cyl,
    data = data,
    family = cumulative("probit")
  )
mdl

avg_predictions(
    mdl,
    by = "cyl",
    type = "prediction"
  ) |> 
  print() |>
  get_draws("DxP") |>
  head()

avg_predictions(
  mdl,
  by = "cyl",
  type = "response"
) |> 
  print() |>
  get_draws("DxP") |>
  head()