In 'new_expression', just parse the constructor as an expression

This avoids the premature parsing decision that happened in the last commit.
However, we now get a conflict because of the '?.' token, for the following
token sequence:

new foo ?.

One one hand, the '?.' token may be part of a member_expression or
subscript_expression. In that case, we want to SHIFT the '?.' token
before creating the 'new_expression' node, in order to group things
like this:

'new' ( expression '?.' identifier)   or
      |                           |
      ----- member_expression ----

On the other hand, the '?.' token may be part of a 'call_expression',
in which case, the parser thinks we would want to create the
'new_expression' first, so as to group things like this:

( 'new' expression ) '?.' arguments
|                  |
-- new_expression --

This second conflicting interpretation arises because 'new_expression'
currently has *higher* precedence than 'call_expression' - we set the
precedences this way in PR #89.

Author: maxbrunsfeld

grammar.js

@@ -632,7 +632,7 @@ module.exports = grammar({
 
     new_expression: $ => prec.right(PREC.NEW, seq(
       'new',
-      field('constructor', $._constructable_expression),
+      field('constructor', $._expression),
       field('arguments', optional($.arguments))
     )),
 

src/grammar.json

@@ -2739,7 +2739,7 @@
             "name": "constructor",
             "content": {
               "type": "SYMBOL",
-              "name": "_constructable_expression"
+              "name": "_expression"
             }
           },
           {