fix(daemon): bound the handshake profile-identity binding on connection serving

[ScriptedAlchemy]

connection_serving.rs:746 and :1468 await bind_authenticated_profile_identity inside a tokio::select! whose only other arm is peer_fully_closed_after_eof(), so the stage is unbounded. It reaches registered_profile_database() → session_runtime_registry() → a cold DaemonSessionRuntimeRegistryV1::open; the 500 ms PROJECT_OPEN_REQUEST_DEADLINE applies only later in portable_project_server_for_request. Under load a contended cold registry open pins the connection instead of yielding the warming refusal.

Symptoms on Test Linux (Elapsed at the 2 s bound): daemon::tests::bootstrap::direct_tool_cache_miss_returns_warming_while_project_opens_in_background, handshake::{daemon_refreshes_once_only_after_generation_change, initialized_ack_preserves_pending_catalog_refresh_notification}, socket::{portable_broker_routes_multiple_closed_invocations…, socket_client_requires_user_storage_scope_without_project, socket_client_routes_multiple_closed_invocations…, user_session_read_bypasses_unregistered_project_route}, all through the same half-close round-trip helper. Isolated, direct_tool_cache_miss… fails differently (bootstrap.rs:3308, no error message) because it blocks WriterScope::Daemon while a project open takes a project-scoped writer, so its blocking lever is stale.

Fix: apply the same deadline to the binding stage (or move it inside the deadline), then re-point the test at a lever that gates a cold open. Diagnosed 2026-09-05 (hauler cc-5830).

[ScriptedAlchemy]

Fixed on PR #707 at 494aeae: the binding now runs on a detached task under a dedicated 10 s PROFILE_IDENTITY_BIND_DEADLINE (a cold profile open measured 4 ms warm, 331 ms uncontended and over 500 ms with six concurrent openers, so the 500 ms project-open deadline was inside the normal range) and the unix broker reads the first request before binding so the refusal is a typed frame on that request's id. The six named tests pass (hauler cc-5903, 86 passed).