Investigate preceding-sibling::* misuse more broadly

[MichaelChirico]

#2402 exposed an issue about preceding-sibling::*[1] including COMMENT by mistake. Took a quick look for similar such bugs and immediately found one in keyword_quote_linter():

lint("a$ # comment\n'b'", keyword_quote_linter())
# <text>:2:1: warning: [keyword_quote_linter] Only quote targets of extraction with # comment if necessary, i.e., if the name is not a valid R symbol (see ?make.names). Use backticks to create non-syntactic names, or use slot() to extract by string.
# 'b'
# ^~~

# comment in the lint message is supposed to be $.

Filing this issue to fix this and other such instances throughout the codebase. Since they've gone unnoticed thus far, they're mainly about cases where comments go in "unusual" places, but it can't hurt to be more robust.

[MichaelChirico]

Same for preceding-sibling::expr/following-sibling::expr which might unintentionally skip equal_assign or expr_or_assign_or_help expr-ish nodes, a la #2444.

[MichaelChirico]

I think the way forward here is #2191 -- once we get some infrastructure set up for "jiggling" tests, we can do a "fuzz" where we insert COMMENT nodes at random places in the AST.