feat(translation): pooled Interpreter session + manual sweep resumes after drain

- interpreter_client: shared requests.Session with a connect-only retry
  (connect=2, no status retries — we map 429/5xx ourselves) so a proxy reload
  is smoothed and the keep-alive connection is pooled across the sweep.
- translate_posts: on an interrupt (drain), re-enqueue after the Retry-After
  hint / default backoff instead of waiting for the daily beat; self-terminates
  via the health gate. Steady-state one-chunk-per-run on success is unchanged.

Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

tests/test_interpreter_client.py

@@ -30,7 +30,7 @@ def test_translate_maps_batch_in_order(monkeypatch):
             "interpreter": {"engine": "llm", "engine_version": "ollama:x:12b"},
         })
 
-    monkeypatch.setattr(ic.requests, "post", fake_post)
+    monkeypatch.setattr(ic.session, "post", fake_post)
     out = ic.translate(["ねこが可愛い", "金髪ギャル！"], base_url="http://i.lan")
     assert out["translations"] == ["The cat is cute", "Blonde gal!"]
     assert out["detected_lang"] == "ja"
@@ -41,7 +41,7 @@ def test_translate_maps_batch_in_order(monkeypatch):
 
 def test_translate_passthrough_unchanged(monkeypatch):
     # Already-English items come back unchanged in their slot (engine "none").
-    monkeypatch.setattr(ic.requests, "post", lambda *a, **k: _Resp(200, {
+    monkeypatch.setattr(ic.session, "post", lambda *a, **k: _Resp(200, {
         "translatedText": ["already english"],
         "detectedLanguage": {"language": "en"},
         "interpreter": {"engine": "none", "engine_version": None},
@@ -53,7 +53,7 @@ def test_translate_passthrough_unchanged(monkeypatch):
 
 
 def test_translate_503_raises_unavailable(monkeypatch):
-    monkeypatch.setattr(ic.requests, "post", lambda *a, **k: _Resp(503, {}))
+    monkeypatch.setattr(ic.session, "post", lambda *a, **k: _Resp(503, {}))
     with pytest.raises(ic.InterpreterUnavailable):
         ic.translate(["x"], base_url="http://i.lan")
 
@@ -62,7 +62,7 @@ def test_translate_connection_error_raises_unavailable(monkeypatch):
     def boom(*a, **k):
         raise ic.requests.ConnectionError("refused")
 
-    monkeypatch.setattr(ic.requests, "post", boom)
+    monkeypatch.setattr(ic.session, "post", boom)
     with pytest.raises(ic.InterpreterUnavailable):
         ic.translate(["x"], base_url="http://i.lan")
 
@@ -71,7 +71,7 @@ def test_translate_connection_error_raises_unavailable(monkeypatch):
 def test_translate_429_and_5xx_raise_unavailable(monkeypatch, code):
     # A gracefully-draining service behind a reverse proxy returns 429/502/503/504
     # — all mean "retry later", not an opaque error.
-    monkeypatch.setattr(ic.requests, "post", lambda *a, **k: _Resp(code, {}))
+    monkeypatch.setattr(ic.session, "post", lambda *a, **k: _Resp(code, {}))
     with pytest.raises(ic.InterpreterUnavailable) as ei:
         ic.translate(["x"], base_url="http://i.lan")
     assert ei.value.retry_after is None
@@ -79,7 +79,7 @@ def test_translate_429_and_5xx_raise_unavailable(monkeypatch, code):
 
 def test_translate_honours_retry_after_seconds(monkeypatch):
     monkeypatch.setattr(
-        ic.requests, "post",
+        ic.session, "post",
         lambda *a, **k: _Resp(503, {}, headers={"Retry-After": "42"}),
     )
     with pytest.raises(ic.InterpreterUnavailable) as ei:
@@ -90,7 +90,7 @@ def test_translate_honours_retry_after_seconds(monkeypatch):
 def test_translate_retry_after_past_http_date_clamps_to_zero(monkeypatch):
     # HTTP-date form; a past date → non-negative clamp to 0 (deterministic).
     monkeypatch.setattr(
-        ic.requests, "post",
+        ic.session, "post",
         lambda *a, **k: _Resp(
             503, {}, headers={"Retry-After": "Wed, 21 Oct 2015 07:28:00 GMT"}),
     )
@@ -101,7 +101,7 @@ def test_translate_retry_after_past_http_date_clamps_to_zero(monkeypatch):
 
 def test_translate_400_raises_bad_request(monkeypatch):
     monkeypatch.setattr(
-        ic.requests, "post", lambda *a, **k: _Resp(400, {"error": "bad target"})
+        ic.session, "post", lambda *a, **k: _Resp(400, {"error": "bad target"})
     )
     with pytest.raises(ic.InterpreterBadRequest):
         ic.translate(["x"], base_url="http://i.lan")
@@ -111,20 +111,20 @@ def test_translate_empty_is_noop(monkeypatch):
     def boom(*a, **k):
         raise AssertionError("should not call the service for an empty batch")
 
-    monkeypatch.setattr(ic.requests, "post", boom)
+    monkeypatch.setattr(ic.session, "post", boom)
     assert ic.translate([], base_url="http://i.lan")["translations"] == []
 
 
 def test_health_true_when_llm_up(monkeypatch):
     monkeypatch.setattr(
-        ic.requests, "get", lambda *a, **k: _Resp(200, {"engines": {"llm": True}})
+        ic.session, "get", lambda *a, **k: _Resp(200, {"engines": {"llm": True}})
     )
     assert ic.health("http://i.lan") is True
 
 
 def test_health_false_when_engine_down(monkeypatch):
     monkeypatch.setattr(
-        ic.requests, "get", lambda *a, **k: _Resp(200, {"engines": {"llm": False}})
+        ic.session, "get", lambda *a, **k: _Resp(200, {"engines": {"llm": False}})
     )
     assert ic.health("http://i.lan") is False
 
@@ -133,7 +133,7 @@ def test_health_false_on_error(monkeypatch):
     def boom(*a, **k):
         raise ic.requests.ConnectionError("refused")
 
-    monkeypatch.setattr(ic.requests, "get", boom)
+    monkeypatch.setattr(ic.session, "get", boom)
     assert ic.health("http://i.lan") is False