Cap action item dedup context from 50 to 10 items

[beastoin]

Parent: #4635

Problem

extract_action_items() fetches up to 50 recent action items (past 2 days) for deduplication context. Each item includes description, due date, and completion status — adding thousands of input tokens per call.

File: backend/utils/conversations/process_conversation.py:87-91

existing_action_items = action_items_db.get_action_items(uid=uid, start_date=two_days_ago, limit=50)

In prompt (conversation_processing.py:290-302):

existing_items_context = f"\nEXISTING ACTION ITEMS FROM PAST 2 DAYS ({len(items_list)} items):\n"

Proposed Change

1. Reduce limit from 50 to 10 (most recent)
2. Or reduce time window from 2 days to 12 hours
3. Only include pending (non-completed) items — completed items don't need dedup

Impact

• Reduces input tokens by ~500-2000 per call depending on user's action item volume
• Estimated 2-3% reduction in total costs
• Active users with many action items benefit most

Risk

Low — duplicate action items may occasionally slip through, but the LLM already handles near-duplicate detection. 10 recent items should catch the vast majority of duplicates.

[beastoin]

📝 The proposed solutions here are based on code analysis and have not been verified by the CTO. If you have a better approach to address this, feel free to propose it.