feat(telemetry): adopt shared PostHog schema v2 (server events, features, flush)

[jrosskopf]

Summary

Upgrades flapi onto the shared DataZoo telemetry schema (telemetry_schema: 2). flapi is a long-running server, so install_kind="server" and one $session_id per uptime.

• Upgrade the posthog-telemetry submodule to the v2.0.0 API (SetProduct, CaptureFeature, AssociateGroup, CaptureError, Flush).
• Boot: SetProduct("flapi", …), install_kind="server", deployment group (optional account via FLAPI_LICENSE_ID), and server_started{endpoint_count, auth_kind} (counts/kinds only).
• Per request/tool/auth: rest_endpoint_served{method, route_template, status_class, duration_ms, cache_hit} at the single REST choke point; mcp_tool_called{tool, status_class, duration_ms}; auth_enforced{auth_kind, outcome}; $exception with an enumerated error_class.
• Privacy: the route template (never the filled path), and bounded enum/number props only — no URL/query/body/headers/SQL. Verified by a no-leak test that drives the real library transport and asserts the serialized batch is clean.
• Lifecycle: Flush() on SIGINT/SIGTERM and clean exit (the library's at-exit handler discards by design). Single opt-out via --no-telemetry / FLAPI_NO_TELEMETRY / DATAZOO_DISABLE_TELEMETRY / telemetry.enabled: false; optional telemetry.sample_rate for high-QPS.
• Docs: TELEMETRY.md listing exactly what is collected.

cache_hit reports whether the endpoint is cache-backed (flapi has no per-request hit/miss signal — documented in TELEMETRY.md).

Test plan

• Debug build clean; flapi-lib + tests link against the upgraded lib.
• Telemetry unit tests: 12 cases / 65 assertions pass, incl. the real-transport no-leak test (asserts route template present; no filled path / SQL / Authorization / password).
• Smoke (first pass): server boots, requests flow the instrumented choke point, SIGTERM flushes + clean exit.

---

^{Need help on this PR? Tag /codesmith with what you need. Autofix is disabled.}