Enhance logging and robustness; bump vlm-engine

Refactor formatting and improve diagnostics across the Haven VLM connector and engine. Update dependency: vlm-engine to 0.9.6. Add detailed debug/progress logs (semaphore acquisition, task lifecycle, HTTP/VLM timing, hypothesis debug tags), switch to asyncio.as_completed for concurrent task handling, and include FPS/performance logging by exposing get_performance_stats() from the engine. Harden config parsing and validation (validate multiplexer endpoints for max_concurrent, safer defaults for return_confidence), improve error handling and cleanup (engine shutdown), and apply numerous whitespace/quote/style fixes for consistency.

Author: HavenCTO