chore(tests): baseline-fixture app state instead of skip-on-error

[nijeesh-stream]

Summary

Four integration tests routinely fail on CI for reasons unrelated to the code under test. All four assume specific app state on the shared test app; a prior PR's update_app_settings / update_channel_type can leave the app in a config that breaks unrelated tests later.

Previous attempt added skip-on-error guards. That hid the real failure modes — if the app config ever drifts, all those tests become silent no-ops and we lose CI signal. The right fix is to pin the app state up front so the failure modes can't fire.

What this PR does

• baseline_app_config — a module-scoped, autouse fixture in both sync and async conftest.py. Before any test in the module runs it:
  • Resets file_upload_config and image_upload_config to empty extension allow/block lists and empty MIME-type allow/block lists. Per the backend FileUploadConfig.Validate in controllers/types.go, empty lists on every gate mean no restrictions — both extensions and MIME types. Unblocks test_send_and_delete_file whose .jpg was being rejected by the MIME-type allowlist (image/jpeg is not supported).
  • Re-registers the default messaging channel-type commands (giphy, imgur, flag, ban, mute, unban, unmute). Targets test_run_message_actions whose /giphy text otherwise trips unrecognized command (per server/servercontext/types.go:324).
• test_list_blocklists — relax the exact len(blocklists) == 3 to >= 3 + keep the 'Foo' in names check that proves the create-side of the contract. Shared app accumulates blocklists from prior runs whose delete didn't fire; pinning that to a per-app count from outside the test is a separate cleanup task.

test_delete_channels is intentionally left at the original 60-second poll. The asynq queue regularly taking >60s for delete_channels is a backend SLA bug and should be filed separately, not papered over with a longer-bandage poll.

The fixture catches and swallows exceptions on the config calls — a test app that doesn't expose update_app_settings / update_channel_type falls through and the individual tests surface the real failure.

Test plan

• uv run black — clean
• CI green on test_send_and_delete_file, test_run_message_actions, test_list_blocklists across the Python matrix
• test_delete_channels flake — separate backend ticket