There are various other regexes here that are trying to capture tokens
in different contexts but at the very least we should also always be
greedily capturing a series of letters, numbers and underscores.
It's OK if this is already covered in some cases by another regex since
we de-duplicate tokens anyway.

The test included in this change is an example where we don't correctly
capture this token today and it is a common example in Ruby so we should
cover it.