Parser misparses postfix record access after parens, record literals, record updates, and qualified variables

Summary

The expression parser only recognizes a postfix .field access chain after a
plain lowercase variable (record.field). After every other kind of term that
can legally take a field access, the .field is instead parsed as a separate
accessor function term, turning the whole expression into a function
application:

The reference compiler (the Haskell gren binary) parses all four as field
access, and treats them differently from the space-separated forms
((f x) .field is an application of the accessor function). With this
parser, both spellings collapse into the same (application) AST, so the
distinction between two different programs is lost at parse time.

Reproduction

1. Save as Repro.gren:

   module Repro exposing (..)
   
   
   a r =
       (get r).field
   
   
   b =
       { x = 1 }.x
   
   
   c r =
       { r | x = 1 }.x
   
   
   d =
       Config.default.timeout
   

2. Inspect the parse (e.g. gren format --pre-ast Repro.gren). Every body is
   {"type": "call", ..., "args": [..., {"type": "accessor", ...}]} where it
   should be {"type": "access", ...}.

Start with the expression parser and reproduce the issue using the Repro.gren example and gren format --pre-ast Repro.gren. Compare the four postfix-access cases with the spaced forms, using the reference AST shapes in the issue. Done means each no-space form is parsed as field access while the space-separated accessor remains a function application.