v1.3.148

Performance

verify-ci parallelization: Relaxed overly conservative -p 1 -parallel 1 flags to default Go parallelism. Reduced CI time from ~3min to ~1min on 64GB machines.
Harness timeout test: Made ClassifyWithLLM timeout configurable via classifyTimeout package var. Test now uses 200ms instead of 8s hardcode, saving 7.8s per run.

429 rate limit retry: Added retryOn429 for Slack, Discord, and Mattermost adapters. Exponential backoff with server-provided retry_after support, max 3 retries.
API token error propagation: apiRequest now detects token-invalid errors in response body and returns ErrInvalidToken instead of silently swallowing the error.
Binding watcher hardening: Auto-mutes stale ownership bindings, skips already-muted bindings, no-mute on same session, restarts on rebind.
LanChat rate limiting: send (5/3s), broadcast (3/3s), send_team (3/3s). Human identity not limited.

Onboarding wizard improvements and i18n updates (de, es, fr, ja, ko, pt, ru, vi, zh-TW).
Desktop config and app binding updates.