fix: Optimize thought completion marking from O(n²) to O(n)

[charlesvien]

Problem

markThoughtCompletion scanned all subsequent items for every thought item using .slice().some(), making it O(n²) in conversation length.

Changes

1. Replace nested scan with a single reverse pass using a Set to track seen turn contexts
2. Reduce time complexity from O(n²) to O(n) by eliminating repeated .slice().some() calls
3. Preserve identical completion semantics (checks for subsequent session_update or turnComplete)

How did you test this?

Manually

[charlesvien]

• fix: Optimize thought completion marking from O(n²) to O(n) #1486 👈 (View in Graphite)
• main

This stack of pull requests is managed by Graphite. Learn more about stacking.