[atomantic]

Summary

• Adds the OpenCode llama TUI (opencode-llama-tui) enabled provider preset for local llama-server instances.
• Enables speculative decoding support with DFlash 2 drafter models (such as z-lab/Muse-Glimmer-30B-DFlash2-GGUF and incoai/Qwen3.8-27B-DFlash2-GGUF).
• Surfaces speculative decoding, preset configurations, and UI management for installing, starting, and stopping llama-server on the Local LLMs settings tab and AI Providers page.
• Adds migration 280-opencode-llama-tui-provider.js to seed the enabled provider to existing installations.
• Adds feature documentation in docs/features/dflash2.md.

Verification

• Unit test suite for migration 280, OpenCode configuration, provider model prefixing, model fetcher registry, and llamaServerManager.
• Full server test suite (30,634 tests across 1,482 files) passed.
• Full client test suite (8,390 tests across 679 files) and client lint passed.