benchmark-lsp: make testing more consistent/fair

Some linters, such as ESLint, run in two phases: parse, then lint.
'let x, x;' causes some linters to fail during the first phase, meaning
the second phase doesn't run. This makes those linters look faster than
they actually are.

Fix the error-producing pattern used in our benchmarks to activate only
during the last phase of multi-phase linters.

On my fast Linux machine, this changes
eslint-server/open-wait-close/edex-ui-filesystem.class.js from
44.3~50.8 ms to 69.9~83.4 ms.

Author: strager

benchmark/benchmark-lsp/corpus/edex-ui-filesystem.class.js

@@ -740,4 +740,4 @@ class FilesystemDisplay {
 module.exports = {
     FilesystemDisplay
 };
-let errorPlease, errorPlease; // Intentional error to force diagnostics.
+if (false) undeclaredVariable.hasOwnProperty(); // Intentional error to force diagnostics.

benchmark/benchmark-lsp/corpus/express-router.js

@@ -660,4 +660,4 @@ function wrap(old, fn) {
     fn.apply(this, args);
   };
 }
-let errorPlease, errorPlease; // Intentional error to force diagnostics.
+if (false) undeclaredVariable.hasOwnProperty(); // Intentional error to force diagnostics.

benchmark/benchmark-lsp/corpus/tiny.js

@@ -1 +1 @@
-let x, x;
+if (false) undeclaredVariable.hasOwnProperty();