Refactor Sentry.close (#2207)

When reviewing #2206, I spotted 2 things that could be improved:

- When closing the SDK, we shouldn't need to enqueue the last client report
  sending to background worker again. This makes the behaviour of `close`
  more difficult to predict.
- Since the session flusher can still enqueue a job to background worker
  when flushing, we should postpone background worker's shutdown instead.

sentry-ruby/lib/sentry-ruby.rb

@@ -233,13 +233,8 @@ def init(&block)
     #
     # @return [void]
     def close
-      @background_worker.perform { get_current_client&.transport&.flush }
-      # session_flusher internally queues to the background worker too
-      @session_flusher&.flush
-
-      @background_worker.shutdown
-
       if @session_flusher
+        @session_flusher.flush
         @session_flusher.kill
         @session_flusher = nil
       end
@@ -249,10 +244,16 @@ def close
         @backpressure_monitor = nil
       end
 
-      if configuration&.include_local_variables
-        exception_locals_tp.disable
+      if client = get_current_client
+        client.transport.flush
+
+        if client.configuration.include_local_variables
+          exception_locals_tp.disable
+        end
       end
 
+      @background_worker.shutdown
+
       @main_hub = nil
       Thread.current.thread_variable_set(THREAD_LOCAL, nil)
     end