Support explicit Ollama model runs

[solusops]

What changed

• Added explicit backend selection for LM Studio and Ollama.
• Removed the tracked model sequence; collaborators pass one or more explicit model IDs or tags.
• Fixed every run at a 16,384-token context cap and omit reasoning configuration.
• Restored dashboard batch ETA from the explicit model selection stored in the index.

Why

This lets collaborators run only the models they choose while preserving a comparable context limit across backends.

Validation

• Python compilation for runner and dashboard
• Mocked Ollama tags/chat adapter check, including num_ctx=16384
• Live LM Studio model listing and explicit GPT-OSS launch
• Dashboard batch ETA API check