feat(perf): implement HTTP connection pooling with httpx.AsyncClient singleton and Keep-Alive

- get_http_client: singleton AsyncClient with limits(max_keepalive=10, max_connections=20, keepalive_expiry=60s)
- close_http_client: clean socket pool teardown in FastAPI lifespan shutdown
- refactor dictionary fetchers to reuse pooled client and avoid 5x TLS handshakes per word lookup
- add test_http_connection_pooling (22/22 pytest passed)

tests/test_api.py

@@ -1011,3 +1011,34 @@ def test_books_pdf_multipart_upload():
 
     # Cleanup
     client.delete(f"/api/books/{book_id}")
+
+
+def test_http_connection_pooling():
+    import pytest
+    from app.dictionary import get_http_client, close_http_client, lookup_term
+    import asyncio
+
+    # 1. Singleton returns same instance
+    c1 = get_http_client()
+    c2 = get_http_client()
+    assert c1 is c2
+    assert not c1.is_closed
+
+    # 2. Check connection limits configuration
+    pool = c1._transport._pool
+    assert pool._max_keepalive_connections == 10
+    assert pool._max_connections == 20
+
+    # 3. Lookup using pooled client
+    res = asyncio.run(lookup_term("resilience"))
+    assert res["query"] == "resilience"
+    assert "translation" in res
+
+    # 4. Clean closure on shutdown
+    asyncio.run(close_http_client())
+    assert c1.is_closed
+    # Next call creates a new active client
+    c3 = get_http_client()
+    assert c3 is not c1
+    assert not c3.is_closed
+    asyncio.run(close_http_client())