Fix model download and management issues

[jasielmacedo]

This commit fixes model download issues and implements context-aware AI interactions:

Model Download Improvements:

• Add retry logic with exponential backoff (max 3 retries)
• Implement proper error handling for network issues (ECONNRESET, ETIMEDOUT, etc.)
• Add heartbeat monitoring to detect stalled downloads (2-min timeout)
• Add socket keep-alive to prevent connection drops
• Track active downloads to prevent duplicate downloads
• Provide user-friendly error messages
• Add download cancellation support

Context Awareness Implementation:

• Add PageContext type for current page URL, title, content, and selected text
• Add AIContext type combining page context, browsing history, and bookmarks
• Update GenerateOptions and ChatOptions to accept optional context
• Implement buildContextualSystemPrompt() to inject context into AI prompts
• Update generate() and chat() methods to use contextual prompts
• Add page:getContext IPC handler to capture current page info
• Update IPC handlers to validate and pass context to Ollama service

Technical Details:

• Add error field to PullProgress for better error reporting
• Add retry status to show retry attempts to users
• Use HTTP/HTTPS agents with keep-alive for stable connections
• Truncate page content to 5000 chars to avoid token limits
• Validate all context fields for security

This enables the AI to be aware of:

• Current browsing page (URL, title, content)
• Selected text on the page
• Recent browsing history
• User bookmarks

All changes maintain backward compatibility.