Improve location of some custom operators

[WardBrian]

This (maybe?) closes #1915.

It adds the same test as #1916, plus a:

• Definite fix for :=
  • Just lex this as an operator.
• Definite fix for reified custom indexing (e.g, something like ( .@(;..)<- ))
  • We lex this as DOTOP SEMI DOTDOT ... and detect it in the driver code
• Heuristic fix for other custom indexing
  • When we see a DOTOP in the wild, we assume it is single-indexing without assignment and complete it accordingly. E.g., the code name.@( (* whatever here ... *) will get turned into .@ ( ) by reconstruct_id, even if we'd like to also detect things like ; or <- to disambiguate.
  • This is obviously wrong-in-general, but I think it's more useful than the current behavior, since users are encouraged to always define this variant if using the custom indexing, and they're often near each other. I also don't know if it can be done any better, since truly disambiguating requires I think arbitrary amounts of parsing?

The test shows both the cases that don't work before (see existing output in #1917) and the cases that "work" but produce incorrect locations due to the third bullet point above.

[WardBrian]

Fuzzy CI diff looks all intentional to me

[voodoos]

Thanks for the patch @WardBrian. I will review it when I get some spare cycles (it might take some time..).

[WardBrian]

@voodoos gentle ping -- is there still interest in this being looked at?