P6B: Ship 14B compacted model, research 32B QAT

[joelteply]

Ship Qwen2.5-Coder-14B compacted at Q5_K (fits 16GB Air). Qwen2 chat template, persona config, auto device detection. See P6B.

[joelteply]

Superseded by #417 (Qwen3.5-35B-A3B evaluation). The 14B compacted model is published but the new MoE reasoning model is a better candidate. Q3_K_M at 16.8GB may fit MacBook Air 16GB. Evaluation in progress on 5090 tower.