OpenCode v1.14.42 adds LLM core, Scout

This reads like infrastructure work that will compound: OpenCode is turning from a terminal agent into a more modular platform for model routing, research, and interactive workflows.

• –The new `@opencode-ai/llm` layer suggests the team is decoupling provider logic from product code, which usually pays off in faster model support and fewer edge-case bugs.
• –Scout is the most user-visible addition here; repo research and docs lookup make the agent smarter before it starts editing.
• –Moving HTTP routes onto Effect and tightening auth/validation points to a cleanup phase where reliability matters more than flashy features.
• –The provider retry and reasoning-option fixes are practical for anyone juggling multiple vendors, since model heterogeneity is where coding agents tend to break first.
• –Interactive `opencode run` support pushes the tool closer to a real terminal-first IDE loop, not just a batch command runner.