feat: GAIA Code initialization

[itomek]

Quick Check ✨

• I've taken a look at existing feature requests
• This feature request relates to GAIA UI (Open-WebUI)

What's on your mind?

GAIA code should follow the same CLI path as EMR did by utilizing gaia-code init.

For example:

uv run gaia-emr init
warning: No `requires-python` value found in the workspace. Defaulting to `>=3.12`.

╭─────────────────────────────────────────╮
│ EMR Agent Setup                         │
│ Downloading and loading required models │
╰─────────────────────────────────────────╯

Step 1: Checking Lemonade server...
  ✓ Lemonade server is running
  ✓ Context size: 32,768 tokens (recommended: 32,768)

Step 2: Checking required models...
  ✓ VLM: Qwen3-VL-4B-Instruct-GGUF
  ✓ LLM: Qwen3-Coder-30B-A3B-Instruct-GGUF
  ✓ Embedding: nomic-embed-text-v2-moe-GGUF

Step 3: Loading required models...
  Loading models into memory for fast inference...

  Loading VLM: Qwen3-VL-4B-Instruct-GGUF...
  ✓ VLM loaded (4.0s)
  Loading LLM: Qwen3-Coder-30B-A3B-Instruct-GGUF...
  ✓ LLM loaded (8.2s)
  Loading Embedding: nomic-embed-text-v2-moe-GGUF...
  ✓ Embedding loaded (0.1s)

  Clearing VLM context for clean memory...
[2026-01-08 10:31:00] | INFO | gaia.llm.lemonade_client.unload_model | lemonade_client.py:2275 | Model unloaded successfully: {'message': 'All models unloaded successfully', 'status': 'success'}
  ✓ VLM context cleared

Step 4: Verifying models are ready...
  ✓ VLM: Ready for form extraction
  ✓ LLM: Ready for chat queries
  ✓ Embedding: Ready for search
  ✓ Context size: 32,768 tokens

Step 5: Model inventory...
  VLM Models: Qwen3-VL-4B-Instruct-GGUF
  LLM Models: Qwen2.5-0.5B-Instruct-CPU, Qwen3-1.7B-GGUF, Qwen3-4B-Instruct-2507-GGUF (+2 more)
  Embedding Models: nomic-embed-text-v2-moe-GGUF
  Total models available: 7
....

[kovtcharov]

Triage check (April 2026): the coder feature shipped via PRs #819-#834 (Phases 1-11) and #886 (semantic_search tool). gaia-code CLI is a registered console_scripts entry. Please re-verify whether the original 'initialization' concern (CLI parity with main gaia command) is still relevant — if yes, the spec needs an update against current shipped behavior; if no, this can close.