Add full mock-provider E2E regression suite and workflow

[Xunzhuo]

Goal

Build a complete E2E regression suite with a mock LLM provider server and GitHub workflow coverage for CLI, Gateway, and core user journeys.

Scope

• Mock LLM provider server with deterministic chat, streaming, tool-call, error, retry, and degraded-response scenarios.
• Full CLI regression for setup, chat, memory updates/search, tools, provider config, cron, and recovery paths.
• Gateway regression for supported IM adapters, duplicate delivery, async replies, cron delivery, and failure messaging.
• Workflow that runs the high-signal E2E suite in CI without requiring live provider credentials.
• Clear separation between fast PR gate, full E2E gate, and optional live certification.

Acceptance Criteria

• A reusable mock provider server is available for E2E tests.
• CLI and Gateway core flows have deterministic E2E coverage.
• The workflow runs in GitHub Actions without external LLM credentials.
• Failure output points to the broken surface and scenario.
• Existing release/live certification tests are not used as a substitute for mock-provider E2E regression.