docs: call it the memory log throughout

- docstrings, comments, messages and the README section use "memory log", matching TradingMemoryLog and memory_log_path
- the write, read and settle paths are named for what they do, not by design phase

tests/test_backtest.py

@@ -1,4 +1,4 @@
-"""Backtesting: many single-shot decisions, scored by the decision log.
+"""Backtesting: many single-shot decisions, scored by the memory log.
 
 A run already records its rating and later settles it with realized and alpha
 return against the regional benchmark. A backtest is that machinery over a grid
@@ -76,7 +76,7 @@ def _config(tmp_path):
 
 
 @pytest.mark.unit
-def test_the_live_decision_log_is_never_written(tmp_path):
+def test_the_live_memory_log_is_never_written(tmp_path):
     config = _config(tmp_path)
     result = run_backtest(["NVDA"], ["2026-01-05", "2026-01-12"], config)
 

tests/test_cli_memory_log.py

@@ -1,4 +1,4 @@
-"""The CLI must use the decision log the same way propagate() does.
+"""The CLI must use the memory log the same way propagate() does.
 
 The CLI streams the graph itself instead of calling propagate(), so memory steps
 that lived only in propagate() never ran on the primary entry point: pending
@@ -160,7 +160,7 @@ def _run_cli(monkeypatch, tmp_path, fake):
 
 
 @pytest.mark.unit
-def test_cli_run_uses_the_decision_log_like_propagate(tmp_path, monkeypatch):
+def test_cli_run_uses_the_memory_log_like_propagate(tmp_path, monkeypatch):
     fake = _FakeGraph()
     _run_cli(monkeypatch, tmp_path, fake)