Add context window limit notification system

[yurekami]

Summary

Implements a context window limit notification system to improve user experience during long conversations. This directly addresses issue #1050.

Features Added:

• ContextTracker utility class (inference/context_tracker.py):

  • Real-time token count tracking
  • Visual progress bar with color-coded status (green → yellow → red)
  • Warning notifications at 75% and 90% thresholds
  • Estimates for remaining tokens and conversation turns

• New interactive commands:

  • /context - Display current context window usage
  • /summary - Get a prompt to summarize the conversation
  • Enhanced /clear with confirmation message

• Automatic notifications:

  • Progress bar displayed after each response
  • Warnings when approaching context limits
  • Suggestions for summarization to preserve important context

Example Output

📊 Context window: 16,384 tokens
💡 Commands: /exit (quit), /clear (new chat), /context (show usage), /summary (get summary prompt)

>>> Hello!

[Response...]

[██████████░░░░░░░░░░░░░░░░░░░░] Context: 1,234/16,384 (7.5%)

At 75% capacity:

⚠️  Context window at 75% capacity.
💡 Tip: Consider summarizing key points to preserve important context.

Test plan

• Verify syntax with python -m py_compile inference/context_tracker.py inference/generate.py
• Test interactive mode shows context usage after each response
• Test /context command displays usage stats
• Test /summary command shows summarization prompt
• Test /clear resets the context tracker
• Verify warnings appear at 75% and 90% thresholds

Closes #1050

🤖 Generated with Claude Code