statusLine: total_input_tokens should include system prompt, tools, and memory

Summary

context_window.total_input_tokens in the statusLine JSON doesn't include the full context - it appears to only count user message tokens.

Expected

total_input_tokens should include all input tokens sent to the model:

• System prompt
• System tools
• Memory files
• Messages

Actual

At conversation start, /context shows ~11k tokens used:

System prompt: 2.4k tokens
System tools: 7.9k tokens
Memory files: 578 tokens
Messages: 95 tokens

But total_input_tokens reports a much smaller number (seemingly just messages), making context percentage calculations inaccurate in custom status lines.