test: lift coverage 41% → 76% (+1068 tests across 36 files)

Service-level tests now exercise provider, permissions, project, journal,
messaging, work_session, metrics, kanban, extraction, learning, notification,
dashboard, llm_routing, a2a, task, repository_base, audit, db_seed,
branch_name, indexed_document, query_helpers, agent. API route tests cover
provider, journal, project, sessions, dashboard, work_session, tasks, a2a,
groups, notifications, agents, channels, messages, kanban, api_resources.
Pure-function helpers covered: handlers, deps_helpers, middleware,
middleware_docs, transcription, pr templates, agents_config, errors,
logging, journal/notification/channel/a2a access, task_lifecycle,
streaming, converters, crypto, schemas (common + websocket), events,
permissions extras.

pyproject ruff per-file-ignores extended for tests so PLR2004 (status code
magic values), PLC0415 (lazy imports), PLR0913 (fixture params), ARG001
(unused fixture deps), SIM105, and E501 don't fight test idioms.

pyproject.toml

@@ -162,8 +162,17 @@ select = [
 "roboco/services/*.py" = ["PLC0415"]
 "roboco/api/routes/*.py" = ["PLC0415"]
 "roboco/runtime/*.py" = ["PLC0415"]
-# Test fixtures that reload modules to test env-var-at-import-time behavior
-"tests/unit/mcp_servers/*.py" = ["PLC0415"]
+# Tests freely use magic values (status codes, indices), lazy imports for
+# dependency-isolation, and many fixture parameters — these style rules
+# are noise in test files.
+"tests/**/*.py" = [
+    "PLR2004",  # magic values (status codes, indices) are normal in tests
+    "PLC0415",  # lazy imports for dependency-isolation in tests
+    "PLR0913",  # many fixture parameters are normal in tests
+    "ARG001",   # unused fixture parameters (db_session for autouse, caplog) are normal
+    "SIM105",   # contextlib.suppress vs try/except/pass — both fine in tests
+    "E501",     # long lines in test fixtures/payloads are common
+]
 
 # =============================================================================
 # MyPy Configuration