feat: Context tracker for multi-step retrieval (Neural Mesh RAG Phase 1)

[mrveiss]

Context

Part of Neural Mesh RAG (#1994), Phase 1, Task 1.5.

Problem

No mechanism to prevent redundant chunk reads across multi-step retrieval. When agentic RAG (Phase 5) executes multiple retrieval steps, the same chunks get re-read and re-ranked, wasting tokens and latency.

Implementation

• Create ContextTracker class with per-query-session scope
• filter_unseen(chunks) — return only chunks not seen in this session
• record(chunk_ids, tokens) — mark chunks as seen, track token budget
• summary() — return session stats for logging
• Token budget tracking with tokens_remaining property

Files

• Create: autobot-backend/knowledge/search_components/context_tracker.py
• Test: autobot-backend/knowledge/search_components/context_tracker_test.py

Plan Reference

docs/plans/2026-03-22-neural-mesh-rag-implementation.md Task 1.5

[mrveiss]

Merged in PR #2019. Context tracker for multi-step retrieval. Part of Neural Mesh RAG #1994.