Fix thread affinity in production request pipeline

Consolidate the request pipeline into a single executor call so that
request_started, before-middleware, view dispatch, after-middleware,
and request_finished all run on the same thread. This preserves
thread-local DB connection assumptions (threading.local).

Move request_finished from response.close() into the pipeline so it
fires on the correct thread. The test client now calls the sync
pipeline directly, removing the need for _SameThreadExecutor and
unconditional asyncio.run().

Author: davegaeddert

plain/plain/http/response.py

@@ -14,7 +14,6 @@
 from http.cookies import SimpleCookie
 from typing import IO, Any
 
-from plain import signals
 from plain.http.cookie import sign_cookie_value
 from plain.json import PlainJSONEncoder
 from plain.utils import timezone
@@ -135,9 +134,6 @@ def __init__(
                 "'content_type' parameter is provided."
             )
         self._resource_closers = []
-        # This parameter is set by the handler. It's necessary to preserve the
-        # historical behavior of request_finished.
-        self._handler_class = None
         self.cookies = SimpleCookie()
         self.closed = False
         if status_code is not None:
@@ -320,7 +316,6 @@ def close(self) -> None:
         # Free resources that were still referenced.
         self._resource_closers.clear()
         self.closed = True
-        signals.request_finished.send(sender=self._handler_class)
 
 
 class Response(ResponseBase):