perf(compression): skip durable refresh for in-memory-only blocks

The ineffective-compression counter is not durable; when it is the sole
reason the gate is blocked there is nothing in the DB that could unblock
it, so re-reading the guard rows on every gate check for the rest of the
session is pure waste. Guard test pins the no-DB-touch behavior.

agent/context_compressor.py

@@ -1593,6 +1593,16 @@ class ContextCompressor(ContextEngine):
         # and re-evaluate so a stale local block cannot outlive the durable
         # state that justified it. The unblocked hot path above never pays
         # for the DB reads.
+        if (
+            self._summary_failure_cooldown_until <= time.monotonic()
+            and self._fallback_compression_streak < 2
+        ):
+            # Blocked solely by the in-memory ineffective-compression
+            # counter, which is not durable — there is nothing in the DB
+            # that could unblock it, so skip the refresh (otherwise this
+            # branch would re-read the DB on every gate check for the rest
+            # of the session).
+            return True
         self._refresh_durable_guards()
         return self._automatic_compression_blocked_locally()
 

tests/agent/test_compression_rotation_state.py

@@ -577,3 +577,22 @@ class TestCooldownPersistFailureIsNotAClearedRow:
         db.clear_compression_failure_cooldown(session_id)
         assert compressor.get_active_compression_failure_cooldown(refresh=True) is None
         assert compressor._summary_failure_cooldown_until == 0.0
+
+    def test_ineffective_count_only_block_skips_durable_refresh(
+        self,
+        refresh_state_db: SessionDB,
+    ):
+        """A block owed solely to the in-memory ineffective counter (which is
+        not durable) must not re-read the DB on every gate check."""
+        db = refresh_state_db
+        session_id = "INEFFECTIVE_ONLY_BLOCK"
+        db.create_session(session_id, source="telegram")
+        compressor = _bound_context_compressor(db, session_id)
+        compressor._ineffective_compression_count = 2
+
+        with patch.object(
+            compressor,
+            "_refresh_durable_guards",
+            side_effect=AssertionError("nothing durable to refresh"),
+        ):
+            assert compressor._automatic_compression_blocked() is True