Make facilitate more robust to crazy fdata() data.frames

[lianos]

I had a DGEList with some crazy y$genes data.frame. Had columns named NA and other stuff. It totally screwed things in weird places.

For instance a call to fdge in something like:

y.all %>%
  facilitate() %>%
  flm_def("group", "a", "b") %>%
  fdge()

would throw a crazy error like this:

Error in initialize(...) : attempt to use zero-length variable name
exiting from: biocbox.FacileLinearModelDefinition(x, assay_name, method, features, 
    filter, with_sample_weights = with_sample_weights, weights = weights, 
    ...)
Error in fdge.FacileLinearModelDefinition(., method = "voom") : 
  rownames of design(x) does not match biocbox

and it turned out that crappy column names in the features() we were getting was the problem