Chained do IIFE: use grammar, not rewriter

[helixbass]

@GeoffreyBooth in playing with whether INDENT_SUPPRESSOR could be used for chaining, I came across #4666/#4672, where the consensus was that a grammar-based solution for handling chained do IIFE eg:

do ->
  a
.b

would be better than the rewriter-based one added in those PRs

At this point I was comfortable tackling a grammar-based solution, so this PR removes that rewriter pass and instead adds tagging in the lexer to distinguish DO (not followed by IIFE) from DO_IIFE and grammar rules for both (and assigns higher precedence to DO_IIFE) - this should be more or less what @jashkenas was pushing for

[GeoffreyBooth]

@helixbass Do you want to resolve the conflicts so we can wrap this up?

[helixbass]

@GeoffreyBooth ok merged ast

[GeoffreyBooth]

This looks good to me. @zdenko or @vendethiel or anyone else, any notes?