Improve "word" experience in vscode.

The model (form language-configuration.json) accepts '-' as long as it
is surrounded by '\w'.

However, the editor still uses '-' as word boundary. This combination
sounds counter-intuitive, but works okish with highlighting: the
model is used to find the initial word, and the editor then finds
"whole" words by using the separators. This means that `--foo-bar` finds
other `foo-bar`s (and, more importantly, `foo-bar` also finds
`--foo-bar`). However, it also means that `foo-bar-gee` will now also be
highlighted for the `foo-bar` part.

A correct solution will use the LSP, which can provide a good list of
the identifiers that are relevant. (Initially maybe just all
identifiers that would match).

vscode/language-configuration.json

@@ -28,7 +28,7 @@
         ["\"", "\""],
         ["'", "'"]
     ],
-    "wordPattern": "(-?\\d*\\.\\d\\w*)|([^\\`\\~\\!\\@\\#\\%\\^\\&\\*\\(\\)\\=\\+\\[\\{\\]\\}\\\\\\|\\;\\:\\'\\\"\\,\\.\\<\\>\\/\\?\\s\\$]+)",
+    "wordPattern": "(-?\\d*\\.\\d\\w*)|(\\w(-?\\w)*)",
     "indentationRules": {
         "increaseIndentPattern": "[\\:]([\\s]*\\|(\\s*[\\w-]+)+\\s*\\|)?$",
         "decreaseIndentPattern": "^[\\s]*else[\\:]$"

vscode/package.json

@@ -26,7 +26,6 @@
 	"contributes": {
 		"configurationDefaults": {
 			"[toit]": {
-				"editor.wordSeparators": "`~!@#$%^&*()=+[{]}\\|;:'\",.<>/?",
 				"editor.indentSize": 2,
 				"editor.insertSpaces": true,
 				"editor.detectIndentation": false