Model download UX: no progress indicator and no model picker

[eshork]

Symptom

1. No progress feedback: Starting a model download appears to hang — no progress bar, percentage, speed, or status indicator. User has no way to know if the download is working, stalled, or failed.

2. No model picker: The only way to select a model is to manually type the model name into a text input (which shows an example placeholder). There is no browsable list of available models to choose from.

Expected Behavior

1. Progress indicator: Model downloads should show real-time progress — at minimum a progress bar with percentage and download speed. Ollama's pull API streams progress events that can be consumed for this.

2. Model picker: Present a selectable list of recommended/available models, ideally organized by size and capability (e.g., fits in 8GB VRAM, 12GB, 24GB+). The text input can remain as a fallback for custom model names.

Suggested Fix

1. Use Ollama's streaming pull API (POST /api/pull with stream: true) to get progress updates and render a Textual progress bar widget
2. Add a model selection screen/list that queries Ollama's library or presents a curated list of recommended models with VRAM requirements

[eshork]

Design Decision (from planning session)

Model list source: Both curated list AND Ollama library

• Show a curated list of recommended models first, organized by VRAM tier (e.g., 6GB, 8GB, 12GB, 24GB+)
• Provide an option to browse the full Ollama library
• Keep the text input as a fallback for custom model names

Progress indicator: Use Ollama's streaming pull API (POST /api/pull with stream: true) to get progress events and render a Textual ProgressBar widget with percentage and status.

[eshork]

resolved in #15