feat: Add task-mcp server for task management via MCP

Implements 10 MCP tools for task management:
- CRUD: task_add, task_list, task_show, task_close, task_update
- Search: task_similar (pgvector), task_context
- Relations: task_link, task_checklist_add, task_checklist_toggle

Uses PostgreSQL with pgvector for semantic search via LiteLLM embeddings.
Connects via SSH tunnel to docker-host:5435.

Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>

src/embeddings.ts

@@ -0,0 +1,58 @@
+// Embeddings via LiteLLM API
+
+const LLM_API_URL = process.env.LLM_API_URL || 'https://llm.agiliton.cloud';
+const LLM_API_KEY = process.env.LLM_API_KEY || '';
+
+interface EmbeddingResponse {
+  data: Array<{
+    embedding: number[];
+    index: number;
+  }>;
+  model: string;
+  usage: {
+    prompt_tokens: number;
+    total_tokens: number;
+  };
+}
+
+/**
+ * Generate embedding for text using LiteLLM API
+ */
+export async function getEmbedding(text: string): Promise<number[] | null> {
+  if (!LLM_API_KEY) {
+    console.error('LLM_API_KEY not set, skipping embedding');
+    return null;
+  }
+
+  try {
+    const response = await fetch(`${LLM_API_URL}/v1/embeddings`, {
+      method: 'POST',
+      headers: {
+        'Authorization': `Bearer ${LLM_API_KEY}`,
+        'Content-Type': 'application/json',
+      },
+      body: JSON.stringify({
+        model: 'text-embedding-ada-002',
+        input: text,
+      }),
+    });
+
+    if (!response.ok) {
+      console.error('Embedding API error:', response.status, await response.text());
+      return null;
+    }
+
+    const data = await response.json() as EmbeddingResponse;
+    return data.data?.[0]?.embedding || null;
+  } catch (error) {
+    console.error('Embedding generation failed:', error);
+    return null;
+  }
+}
+
+/**
+ * Format embedding array for PostgreSQL vector type
+ */
+export function formatEmbedding(embedding: number[]): string {
+  return `[${embedding.join(',')}]`;
+}