Fix highlighting after disembodied IPA branch

Fixes #543
```ql
newtype TA = TB()

private predicate foo() { any() }
```
Our TextMate grammar didn't realize that the newtype declaration ended after the closing paren of the branch's parameter list, so the `private` modifier was highlighted incorrectly.

It's surprisingly tricky to get TextMate to handle this correctly, so I wound up just treating the IPA declaration head (`newtype TA`), the branch head (`= TB`), the branch parameter list, and the branch body as directly children of the module body. This is kind of hacky, but it does fix the bug without introducing any new cases where we have incorrect highlighting of valid code.

extensions/ql-vscode/syntaxes/ql.tmLanguage.yml

@@ -99,6 +99,12 @@ repository:
   comment-start:
     match: '// | /\*'
 
+  # A pattern that can start a run of whitespace or a comment.
+  # Commonly used as a negative lookahead in the `end` regex of a nonterminal, when searching for
+  # tokens that can't start a child of that nonterminal.
+  whitespace-or-comment-start:
+    match: '\s | $ | (?#comment-start)'
+
   # All tokens that can appear in any context.
   non-context-sensitive:
     patterns:
@@ -113,7 +119,7 @@ repository:
     name: keyword.operator.relational.ql
   
   comparison-operator:
-    match: '=|\!-'
+    match: '=|\!\='
     name: keyword.operator.comparison.ql
   
   arithmetic-operator:
@@ -610,6 +616,12 @@ repository:
     - include: '#import-directive'
     - include: '#import-as-clause'
     - include: '#module-declaration'
+    - include: '#newtype-declaration'
+    # See the comment on newtype-declaration for why we include these next three nonterminals at the
+    # module-member level instead of as part of the newtype-declaration.
+    - include: '#newtype-branch-name-with-prefix'
+    - include: '#predicate-parameter-list'
+    - include: '#predicate-body'
     - include: '#class-declaration'
     - include: '#select-clause'
     - include: '#predicate-or-field-declaration'
@@ -781,7 +793,7 @@ repository:
   bindingset-annotation:
     beginPattern: '#bindingset'
     # Ends after the next `]`, or when we encounter something other than a `[`.
-    end: '(?! \s | (?#comment-start) | \[ ) |
+    end: '(?! (?#whitespace-or-comment-start) | \[ ) |
           (?<=\])'
     name: meta.block.bindingset-annotation.ql
     patterns:
@@ -802,7 +814,7 @@ repository:
   language-annotation:
     beginPattern: '#language'
     # Ends after the next `]`, or when we encounter something other than a `[`.
-    end: '(?! \s | (?#comment-start) | \[ ) |
+    end: '(?! (?#whitespace-or-comment-start) | \[ ) |
           (?<=\])'
     name: meta.block.language-annotation.ql
     patterns:
@@ -824,7 +836,7 @@ repository:
   pragma-annotation:
     beginPattern: '#pragma'
     # Ends after the next `]`, or when we encounter something other than a `[`.
-    end: '(?! \s | (?#comment-start) | \[ ) |
+    end: '(?! (?#whitespace-or-comment-start) | \[ ) |
           (?<=\])'
     name: meta.block.pragma-annotation.ql
     patterns:
@@ -841,34 +853,53 @@ repository:
       name: storage.modifier.ql
 
   # The declaration of an IPA type.
+  # This only includes the `newtype` keyword and the identifier for the IPA type itself. The
+  # branches of the IPA type are modeled as separate nonterminals contained directly in the module
+  # body. This is kind of hacky, but without it, we don't seem to have a way to get TextMate to
+  # handle this:
+  # ```ql
+  # newtype TRoot =
+  #   TBranch1(int x) {
+  #     x = 5
+  #   } or
+  #   TBranch2()  // No body
+  #
+  # TOther getOther() { any() }
+  # ```
+  # If the branches are within the newtype declaration node, it's very hard to get the upper-id for
+  # the name of the IPA type to be included in the newtype declaration node, without also including
+  # the `TOther` upper-id in the declaration node.
   newtype-declaration:
     beginPattern: '#newtype'
-    # Ends when we see something other than one of:
-    # - An upper-id (branch name)
-    # - A comment
-    # - Whitespace
-    # - `=`
-    # - `(`
-    end: '(?! \s | (?#upper-id) | (?#comment-start) | \= | \( )'
-    name: meta.block.newtype.ql
-    patterns:
-    - include: '#non-context-sensitive'
-    - include: '#newtype-branch'
-
-  # The branch of an IPA type.
-  newtype-branch:
-    begin: '(?#upper-id)'
-    beginCaptures:
+    # We're expecting a newtype-declaration-without-keyword immediately after the `newtype` keyword,
+    # so end if we see anything other than the upper-id that starts it, or whitespace, or a comment.
+    # An upper-id can't start anything else at module scope, so once we see the rest of this
+    # newtype-declaration, whatever comes next should end this block.
+    end: '(?#upper-id)'
+    endCaptures:
       '0':
         name: entity.name.type.ql
-    # Ends after a `}`, or when we encounter something other than a `{`.
-    end: '(?<=\}) | (?! \s | (?#comment-start) | \{ )'
-    name: meta.block.newtype-branch.ql
+    name: meta.block.newtype-declaration.ql
+    patterns:
+    - include: '#non-context-sensitive'
+
+  # A branch of an IPA type, including just the `=` or `or` prefix and the name of the branch.
+  # The parameter list and body are separate nonterminals contained directly within the module body.
+  # See the comment for newtype-declaration for why.
+  newtype-branch-name-with-prefix:
+    begin: '\= | (?#or)'
+    beginCaptures:
+      '0':
+        patterns:
+        - include: '#or'
+        - include: '#comparison-operator'
+    end: '(?#upper-id)'
+    endCaptures:
+      '0':
+        name: entity.name.type.ql
+    name: meta.block.newtype-branch-name-with-prefix.ql
     patterns:
-    - include: '#predicate-body'
     - include: '#non-context-sensitive'
-    - match: '(?#upper-id)'
-      name: entity.name.type.ql
 
   # The declaration of a class, include an alias.
   class-declaration: