Use PostHog SDK for self-hosted telemetry

[suguanYang]

Summary

• replace the raw PostHog /batch HTTP transport with the official posthog Python SDK behind the existing TelemetryClient wrapper
• keep Knowhere's bounded async queue, event-name allowlist, property sanitizer, stable installation distinct_id, $process_person_profile=false, and disable_geoip=True
• add anonymous self-hosted aggregate telemetry for usage, retrieval, worker, API, and provider activity:
  • self_hosted_usage_aggregate
  • self_hosted_retrieval_aggregate
  • self_hosted_worker_aggregate
  • self_hosted_api_aggregate
  • self_hosted_provider_aggregate
• collect only aggregate counters/latency buckets and keep raw user IDs, document names, queries, prompts, responses, keys, and model names out of event payloads
• add bounded process-local API request metrics and DB-backed aggregate collectors that continue to fail closed if telemetry collection fails
• add SDK and aggregate-focused contract tests using a fake PostHog client instead of asserting raw HTTP payloads

Verification

• uv run pytest apps/api/tests/contract/test_self_hosted_telemetry_contract.py -q
• uv run pytest apps/api/tests/contract/test_s3_event_contract.py apps/api/tests/contract/test_self_hosted_telemetry_contract.py -q
• make lint
• make typecheck
• git diff --check
• Built knowhere-self-hosted:telemetry-aggregates from this runtime branch plus the self-hosted PR source bundle, started the compose stack, verified /health and dashboard /login, registered a guest, created/uploaded a text job through LocalStack SNS, waited for the job to complete, waited for the next aggregate interval, and stopped the app to flush telemetry.