LibJS: Be more lenient when parsing milliseconds for Date

Other engines don't give NaN if there is at least one digit after the
dot for milliseconds. We were much stricter and required exactly three
digits.
But there is real world usage of different amounts of digits such as
discord having three extra trailing zeros.

Userland/Libraries/LibJS/Runtime/DateConstructor.cpp

@@ -79,7 +79,33 @@ static Value parse_simplified_iso8601(GlobalObject& global_object, const String&
         return false;
     };
     auto lex_seconds = [&]() { return lex_n_digits(2, seconds) && *seconds >= 0 && *seconds <= 59; };
-    auto lex_milliseconds = [&]() { return lex_n_digits(3, milliseconds); };
+    auto lex_milliseconds = [&]() {
+        // Date.parse() is allowed to accept an arbitrary number of implementation-defined formats.
+        // Milliseconds are parsed slightly different as other engines allow effectively any number of digits here.
+        // We require at least one digit and only use the first three.
+
+        auto digits_read = 0;
+        int result = 0;
+        while (!lexer.is_eof() && is_ascii_digit(lexer.peek())) {
+            char ch = lexer.consume();
+            if (digits_read < 3)
+                result = 10 * result + ch - '0';
+
+            ++digits_read;
+        }
+
+        if (digits_read == 0)
+            return false;
+
+        // If we got less than three digits pretend we have trailing zeros.
+        while (digits_read < 3) {
+            result *= 10;
+            ++digits_read;
+        }
+
+        milliseconds = result;
+        return true;
+    };
     auto lex_seconds_milliseconds = [&]() { return lex_seconds() && (!lexer.consume_specific('.') || lex_milliseconds()); };
     auto lex_timezone = [&]() {
         if (lexer.consume_specific('+')) {