Always shutdown webrick even if thread is not alive

This should properly shutdown webrick so that we don't need to
forcefully kill the thread to shutdown webrick in the tests.

Without properly shutting down webrick, we leak threads and other
resources to each test cases, eventually causing everything to slow down
and timing out the CI jobs.

This is discovered in
https://gitlab.com/gitlab-org/gitlab/-/issues/462928#note_1916591506

A quick fix was proposed in the above thread and merged in
https://gitlab.com/gitlab-org/gitlab/-/merge_requests/153887

This should properly shutdown webrick instead.

* It should be safe to shutdown webrick regardless
* In a test, we use a thread to wrap around `WEBrick::HTTPServer`,
  turning the blocking server into a non-blocking server. This means
  the daemon thread will stop immediately because everything is
  wrapped inside another thread. If we only shutdown webrick when
  the thread is alive, we're not shutting it down in this test case
  either, leaking resources.

This preserves similar performance comparing to killing the thread.

This partially reverts
https://gitlab.com/gitlab-org/gitlab/-/commit/f5b1317af8aad2d8b7d098c0e4dada25ccfd924b

lib/gitlab/metrics/exporter/base_exporter.rb

@@ -68,11 +68,8 @@ def stop_working
           if server
             # we close sockets if thread is not longer running
             # this happens, when the process forks
-            if thread.alive?
-              server.shutdown
-            else
-              server.listeners.each(&:close)
-            end
+            server.listeners.each(&:close) unless thread.alive?
+            server.shutdown
           end
           @server = nil

spec/lib/gitlab/metrics/exporter/base_exporter_spec.rb

@@ -216,7 +216,7 @@ def call(env)
       # in separate thread
       allow_any_instance_of(::WEBrick::HTTPServer)
         .to receive(:start).and_wrap_original do |m, *args|
-        Thread.new do
+        @server_thread = Thread.new do # rubocop:disable RSpec/InstanceVariable -- let does not work for this case
           m.call(*args)
         rescue IOError
           # is raised as we close listeners
@@ -224,8 +224,17 @@ def call(env)
       end
     end
+    attr_reader :server_thread
+
     after do
       exporter.stop
+
+      next unless server_thread
+
+      server_thread.join(0.05)
+      raise '`exporter.stop` should terminate `server_thread`' if server_thread.alive?
+    ensure
+      server_thread.kill.join if server_thread
     end
     with_them do