Fix QStash webhook delivery state handling

[suguanYang]

Summary

Fixes QStash webhook delivery state handling for retry callbacks and lost terminal callbacks.

• Keep normal QStash callbacks in delivering when the destination response is non-2xx, and only mark them delivered on 2xx responses.
• Preserve terminal webhook event state when an older retry callback arrives late.
• Publish QStash messages with a deterministic deduplication id and label.
• Reconcile stale delivering events from QStash terminal delivery logs during the existing webhook recovery task.
• Add a QStash base URL override for local development and end-to-end verification.

Fixes #24.

Verification

• uv run --group lint ruff check apps/api/app/api/v1/routes/qstash_callbacks.py apps/api/tests/contract/test_qstash_callback_contract.py apps/worker/app/core/tasks/webhook_tasks.py apps/worker/tests/contract/test_webhook_recovery_contract.py packages/shared-python/shared/services/webhook/qstash_publisher.py packages/shared-python/shared/core/config/qstash.py
• uv run --package knowhere-api-app --group dev pytest apps/api/tests/contract/test_qstash_callback_contract.py -q
• uv run --package knowhere-worker-app --group dev pytest apps/worker/tests/contract/test_webhook_recovery_contract.py -q
• Local webhook delivery e2e using the repo local Postgres/Redis stack, the running API QStash callback route, a local QStash-compatible HTTP service, and a local customer webhook returning 503 then 204. Verified callback log statuses [503, 204], final event status delivered, and stale-delivering reconciliation result {"status":"success","recovered":0,"provider":"qstash","reconciled":1}.