Inconsistent behaviour with drupalGet() when site is installed in a subdirectory

Adding more info to the summary.

@catch - it's all a bit tricky but we rely on code in \Drupal\Tests\UiHelperTrait::buildUrl() to add the base directory and language... see:

      // Strip $base_path, if existent.
      $length = strlen($base_path);
      if (substr($path, 0, $length) === $base_path) {
        $path = substr($path, $length);
      }

      $force_internal = isset($options['external']) && $options['external'] == FALSE;
      if (!$force_internal && UrlHelper::isExternal($path)) {
        return Url::fromUri($path, $options)->toString();
      }
      else {
        $uri = $path === '<front>' ? 'base:/' : 'base:/' . $path;
        // Path processing is needed for language prefixing.  Skip it when a
        // path that may look like an external URL is being used as internal.
        $options['path_processing'] = !$force_internal;
        return Url::fromUri($uri, $options)
          ->setAbsolute()
          ->toString();

Changing this behaviour would be extremely tricky.

I think the current different behaviour for sites with and without a base directory is wrong and should be consistent but I think buildUrl should continue to:
1. Add the base directory
2. And do path processing.

I think we could improve things by making it only set the path_processing option if it is not set. Thereby allowing callers to take control of this.

The thing I don't get is why adding

      if (strlen($path) > 1) {
        $path = ltrim($path, '/');
      }

didn't actually fix OpenTelemetryNodePagePerformanceTest::testNodePageHotCache() - v confused about that.

Ah I see why it did not work. The code was added in the wrong place. Fixed on the MR attached here. Let's see if something else then fails... this will be interesting.

Opened a new branch for a test-only change because all the changes are in the test system so the magic test-only job doesn't work.

Tests fail as expected... https://git.drupalcode.org/project/drupal/-/pipelines/146418/test_report...

Updated issue summary slightly to include solution and which MR should be committed

The change of stripping the forward slash seems straight forward.

Test-only MR shows coverage