Add two tier filtering to the knowledge base filter

[ShaohongZ]

Overview

Add two-tier confidence filtering for knowledge base results in the Fuser, with configurable thresholds.

Type of change

• Bug fix
• [v] New feature
• Documentation improvement
• Bounty issue submission
• Other:

Changes

• Two-tier KB filtering: Query results are now split into high confidence (≥ high_score) and low confidence (low_score to high_score) tiers. Results below low_score are discarded. High-confidence results are included directly; low-confidence results are prefixed with [Potentially relevant, low confidence] so the LLM can weigh them appropriately.
• Configurable thresholds: high_score (default 0.92) and low_score (default 0.75) can be set via the knowledge_base config block alongside the existing min_score. min_score is a query-level floor passed to the vector store; high_score/low_score are display-level tiers applied after results are returned.
• Fix duplicate KB query: Removed a duplicate knowledge_base.query() call that was executing the same query twice per tick.
• Updated tests: Fuser tests updated to match the new two-tier format_context call pattern.

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.
✅ All tests successful. No failed tests found.

📢 Thoughts on this report? Let us know!