perf(mcp): Replace redundant LLM usages with deterministic logic

- Refactored memory_consolidation_worker to use purely deterministic string normalization ((N)$ regex stripping & lowercase) instead of passing graph entities to Ollama for duplication detection.
- Removed redundant Ollama architectural summarization from log_code_change tool since the agent already provides a descriptive summary.
- Removed redundant Ollama fix summarization from log_error_fix tool since the agent already provides a solution string.

scratch_schema.rs

@@ -0,0 +1,25 @@
+use schemars::schema_for;
+use schemars::JsonSchema;
+use serde::{Deserialize, Serialize};
+
+#[derive(JsonSchema, Deserialize, Serialize)]
+#[serde(tag = "action", rename_all = "snake_case")]
+pub enum TasksToolEnum {
+    /// Create task.
+    Add {
+        /// Task title.
+        title: String,
+        /// Task description.
+        description: Option<String>,
+    },
+    /// Delete task.
+    Delete {
+        /// Task ID.
+        id: String,
+    }
+}
+
+fn main() {
+    let schema = schema_for!(TasksToolEnum);
+    println!("{}", serde_json::to_string_pretty(&schema).unwrap());
+}