Add no-mock E2E tests for Reminders app (via Gateway app-proxy)

[rowan-stein]

User request

Add proper end-to-end tests for the Reminders app, using real services (no mocks).

Additional requirement (clarification): cover the agent-mediated lifecycle:
thread → agent → (agent schedules reminder) → timer fires → thread gets reminder message → agent wakes + responds.

Specification (research)

Source of truth: agynio/architecture (Centralized E2E + CI integration).

Location + suite

• All E2E tests must live in agynio/e2e (no per-service test/e2e directories).
• Implement in Go under suites/go-core (Go testing + testify/require).

Tagging / selection

• Introduce service tag: svc_reminders (suite + tests).
• Reminders is invoked via Gateway app-proxy and depends on Gateway for delivery; tag the tests to run for either service:
  • //go:build e2e && (svc_reminders || svc_gateway)

“Real” E2E path (no mocks)

• Invoke Reminders through Gateway’s app-proxy (NOT directly hitting the reminders service) to cover:
  • Identity propagation
  • Gateway routing
  • App networking path

Gateway app-proxy contract

• Base URL: AGYN_BASE_URL
• Required headers:
  • Authorization: Bearer <AGYN_API_TOKEN>
  • x-organization-id: <AGYN_ORGANIZATION_ID>
  • Content-Type: application/json (POST)
• Endpoints (via Gateway):
  • GET /apps/reminders/healthz
  • POST /apps/reminders/create-reminder
  • POST /apps/reminders/cancel-reminder
  • POST /apps/reminders/list-reminders
  • POST /apps/reminders/get-reminder

Seeding + validation (real dependencies)

Seed required data via real services (no DB seeding, no mocks):

• Apps gRPC (apps:50051, insecure):
  • AppsService.GetAppBySlug({ slug: \"reminders\" }) → app.identity_id
• Threads gRPC (threads:50051, insecure):
  • ThreadsService.CreateThread({ participant_ids: [...] }) → thread.id
  • ThreadsService.GetMessages({ thread_id, ... }) to assert delivery

Cleanup requirements

• Any test creating reminders must ensure no long-lived pending reminders:
  • Use t.Cleanup to cancel reminders that are not expected to complete quickly.
  • Tests must archive any created threads.
  • Tests must delete any created agents.

Test cases (critical user journeys)

Implement the following E2E tests (all real, no mocks):

1. Gateway→App wiring health: GET /apps/reminders/healthz via Gateway returns 200.
2. Create + Get roundtrip: create reminder (delay ~60s), then get reminder and assert fields are stable; cleanup cancels.
3. List filters: create 2 reminders, cancel 1, validate list default/pending/cancelled/all behavior.
4. Cancel semantics: cancel pending reminder, verify state transition; second cancel returns 409; random UUID returns 404.
5. Delivery happy path: create thread with app participant, create reminder with delay_seconds=0, poll until completed, verify Threads has exactly one message with body Reminder: <note> and sender = appIdentityID.
6. Delivery failure doesn’t mark completed: create reminder for non-existent thread ID, wait briefly, assert still pending and completed_at == null; cleanup cancels.
7. Agent-mediated full loop:
   • Create thread with caller identity + an agent as participants (same org where Reminders is installed).
   • Send a message that triggers the agent to schedule a short reminder (no mocks).
   • Verify sequence:
     1. agent sends a "scheduled" response
     2. reminders app posts Reminder: <note> into the thread
     3. agent sends a follow-up response after the reminder arrives
   • Use a deterministic TestLLM script so the agent reliably executes the scheduling step (shell tool) and produces deterministic messages.

Required e2e repo plumbing

• Update suites/go-core/suite.yaml tags to include svc_reminders.
• Ensure Go suite can generate/use Apps + Threads gRPC clients (update buf.gen.yaml inputs if agynio/api/apps/v1 isn’t currently included).
• Add/extend TestLLM suite definitions as needed so the agent can deterministically run the reminder-scheduling shell command and respond to the reminder message.

[rowan-stein]

New blocker found via real svc_reminders tagged run:

• Gateway app-proxy is dialing apps.v1.App.ziti_service_id as the OpenZiti service name.
• ziti_service_id is the controller’s internal service ID; Apps creates the dialable service name as app-<slug> (e.g. app-reminders).
• This causes service '<id>' not found and 502s from /apps/reminders/*.

Fix needed in agynio/gateway (dial by service name) + then bump bootstrap gateway chart version so CI clusters pick it up.