feat: durable (#794)

* durable (no-mistakes)

* hatchet no-mistake

* hatchet no-mistake

* hatchet no-mistake, better logging

* remove conductor and add hatchet tests (no-mistakes)

* self-review (no-mistakes)

* hatched logs

* remove shadow mode for hatchet

* and add hatchet processor setting to room

* .

* cleanup

* hatchet init db

* self-review (no-mistakes)

* self-review (no-mistakes)

* hatchet: restore zullip report

* self-review round

* self-review round

* self-review round

* dry hatchet with celery

* dry hatched with celery - 2

* self-review round

* more NES instead of str

* self-review wip

* self-review round

* self-review round

* self-review round

* can_replay cancelled

* add forgotten file

* pr autoreviewer fixes

* better log webhook events

* durable_started return

* migration sync

* latest changes feature parity

* migration merge

* pr review

---------

Co-authored-by: Igor Loskutov <igor.loskutoff@gmail.com>

server/tests/conftest.py

@@ -527,6 +527,22 @@ def fake_mp3_upload():
         yield
 
 
+@pytest.fixture(autouse=True)
+def reset_hatchet_client():
+    """Reset HatchetClientManager singleton before and after each test.
+
+    This ensures test isolation - each test starts with a fresh client state.
+    The fixture is autouse=True so it applies to all tests automatically.
+    """
+    from reflector.hatchet.client import HatchetClientManager
+
+    # Reset before test
+    HatchetClientManager.reset()
+    yield
+    # Reset after test to clean up
+    HatchetClientManager.reset()
+
+
 @pytest.fixture
 async def fake_transcript_with_topics(tmpdir, client):
     import shutil