v0.0.5a - vLLM Support + Docs Added

[0.0.5a] - 2026-04-07

Added

• vLLM Engine tab — full start/stop/profiles support mirroring the SGLang tab. Docker-based, with its own profile system (vllm_profiles.json), status LED, 20-second polling, and ready detection requiring both container running + model serving. Status pill in the header bar shows live vLLM health alongside the other services.

• vLLM status pill — header bar now shows four service indicators: SGLang, Ollama, LiteLLM, and vLLM. The vLLM pill displays the active model name when running.

Fixed

• ~/ paths in profile scripts now resolve correctly — both SGLang and vLLM start endpoints now call os.path.expanduser() on the script path before checking existence or executing. Previously, profiles using ~/sglang/start.sh style paths would always fail with "Script not found".

• Docker container detection is now exact-match — status checks for SGLang and vLLM now use name=^sglang$ / name=^vllm$ filters instead of substring matching, preventing false positives from unrelated containers with similar names.

Changed

• config.json — added services.vllm_base field (default http://127.0.0.1:8000).
• docs.html — added vLLM Engine tab guide (section 05), vLLM Profiles reference (section 09), vLLM troubleshooting entries. Architecture diagram updated to three backends. Model recommendation badges made engine-agnostic.
• /api/status — now includes vllm key with ok and model fields.
• /api/nodeinfo — now includes vllm_port.