Add performance features: caching, cost tracking, retry, compaction, classification, scrubbing

Inspired by zeroclaw's lightweight patterns for slow hardware:
- Response cache (SQLite + SHA-256 keyed) to skip redundant LLM calls
- History compaction — LLM-summarize old messages when history exceeds 50
- Query classifier routes simple/research queries to cheaper models
- Credential scrubbing removes secrets from tool output before sending to LLM
- Cost tracker with daily/monthly budget enforcement (SQLite)
- Resilient provider with retry + exponential backoff + fallback provider
- Approval engine gains session "always allow" and audit log

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

xtrm_agent/config.py

@@ -61,6 +61,16 @@ class MCPServerConfig(BaseModel):
     url: str = ""
 
 
+class PerformanceConfig(BaseModel):
+    """Performance tuning — caching, cost tracking, model routing."""
+
+    cache_ttl: int = 3600
+    daily_budget_usd: float = 0.0
+    monthly_budget_usd: float = 0.0
+    fallback_model: str = ""
+    model_routing: dict[str, str] = Field(default_factory=dict)
+
+
 class OrchestratorConfig(BaseModel):
     max_concurrent: int = 5
     delegation_timeout: int = 120
@@ -87,6 +97,7 @@ class Config(BaseModel):
     mcp_servers: dict[str, MCPServerConfig] = Field(default_factory=dict)
     agents: dict[str, str] = Field(default_factory=dict)
     orchestrator: OrchestratorConfig = Field(default_factory=OrchestratorConfig)
+    performance: PerformanceConfig = Field(default_factory=PerformanceConfig)
 
 
 def load_config(path: str | Path = "config.yaml") -> Config: