feat(context): enhance performance, caching, and settings management

- Replace hard-coded limits with configurable settings (e.g., cache memory size, truncation strategy, relevance settings).
- Optimize parallel execution in token counting, scoring, and reranking for source diversity.
- Improve caching logic:
  - Add per-context locks for safe parallel scoring.
  - Reuse precomputed fingerprints for cache efficiency.
- Make truncation, scoring, and ranker behaviors fully configurable via settings.
- Add support for middle truncation, context hash-based hashing, and dynamic token limiting.
- Refactor methods for scalability and better error handling.

Tests: Updated all affected components with additional test cases.

backend/app/services/context/scoring/relevance.py

@@ -9,6 +9,7 @@ import logging
 import re
 from typing import TYPE_CHECKING, Any
 
+from ..config import ContextSettings, get_context_settings
 from ..types import BaseContext, KnowledgeContext
 from .base import BaseScorer
 
@@ -32,7 +33,9 @@ class RelevanceScorer(BaseScorer):
         self,
         mcp_manager: "MCPClientManager | None" = None,
         weight: float = 1.0,
-        keyword_fallback_weight: float = 0.5,
+        keyword_fallback_weight: float | None = None,
+        semantic_max_chars: int | None = None,
+        settings: ContextSettings | None = None,
     ) -> None:
         """
         Initialize relevance scorer.
@@ -40,11 +43,25 @@ class RelevanceScorer(BaseScorer):
         Args:
             mcp_manager: MCP manager for Knowledge Base calls
             weight: Scorer weight for composite scoring
-            keyword_fallback_weight: Max score for keyword-based fallback
+            keyword_fallback_weight: Max score for keyword-based fallback (overrides settings)
+            semantic_max_chars: Max content length for semantic similarity (overrides settings)
+            settings: Context settings (uses global if None)
         """
         super().__init__(weight)
+        self._settings = settings or get_context_settings()
         self._mcp = mcp_manager
-        self._keyword_fallback_weight = keyword_fallback_weight
+
+        # Use provided values or fall back to settings
+        self._keyword_fallback_weight = (
+            keyword_fallback_weight
+            if keyword_fallback_weight is not None
+            else self._settings.relevance_keyword_fallback_weight
+        )
+        self._semantic_max_chars = (
+            semantic_max_chars
+            if semantic_max_chars is not None
+            else self._settings.relevance_semantic_max_chars
+        )
 
     def set_mcp_manager(self, mcp_manager: "MCPClientManager") -> None:
         """Set MCP manager for semantic scoring."""
@@ -112,11 +129,11 @@ class RelevanceScorer(BaseScorer):
                 tool="compute_similarity",
                 args={
                     "text1": query,
-                    "text2": context.content[:2000],  # Limit content length
+                    "text2": context.content[: self._semantic_max_chars],  # Limit content length
                 },
             )
 
-            if result.success and result.data:
+            if result.success and isinstance(result.data, dict):
                 similarity = result.data.get("similarity")
                 if similarity is not None:
                     return self.normalize_score(float(similarity))
@@ -171,7 +188,7 @@ class RelevanceScorer(BaseScorer):
         **kwargs: Any,
     ) -> list[float]:
         """
-        Score multiple contexts.
+        Score multiple contexts in parallel.
 
         Args:
             contexts: Contexts to score
@@ -181,8 +198,10 @@ class RelevanceScorer(BaseScorer):
         Returns:
             List of scores (same order as input)
         """
-        scores = []
-        for context in contexts:
-            score = await self.score(context, query, **kwargs)
-            scores.append(score)
-        return scores
+        import asyncio
+
+        if not contexts:
+            return []
+
+        tasks = [self.score(context, query, **kwargs) for context in contexts]
+        return await asyncio.gather(*tasks)