Add decision models (d1-omni, Laya, basal) #1563

Problem description

Picking from fixed answers with an LLM means generating and parsing text.

Proposed solution

Add a decision task returning probabilities over given answers in one pass:

• LiquidAI/d1-omni-600M
• convaiinnovations/laya (421M, Apache 2.0)
• basal-1.5 nano (1.5B, Bielik-based)

Benefits to React Native ExecuTorch

Fast, calibrated classification and routing.