Generally I have been good about safely accessing the tokens but failed
to properly guard against no tokens in places
where it could theoretically happen through invalid syntax.
I added a test case for one occurance, other changes are theoretical only.
https://github.com/ruby/prism/commit/4a3866af19