fix(live): ROT02/TSM01 scartano la barra 1d in formazione (forming-bar in _panel)

Code-review follow-up: i worker daily valutavano momentum/regime e bookavano il
return sulla candela 1d PARZIALE al primo poll dopo mezzanotte UTC, poi last_bar_ts
bloccava la rivalutazione a giorno chiuso (stessa classe del bug gia' fixato su
fade/TR01/Pairs). Fix in _panel (un punto solo -> ROT02 e TSM01 lo ereditano) via
src.live.bars.last_bar_is_forming. Replay honest invariato (dati storici: nessuna
barra in formazione). Test: 88/88 + parity drop forming daily bar.

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

tests/portfolio/test_rotation_worker.py

@@ -47,6 +47,22 @@ def test_rotation_warns_once_on_short_panel(tmp_path, monkeypatch):
     assert w._panel_warned is False
 
 
+def test_panel_drops_forming_daily_bar(tmp_path):
+    # serie 1d che termina ADESSO -> l'ultima riga e' la candela 1d in formazione
+    import time
+    from src.live.rotation_worker import _panel
+    n = 200
+    now = int(time.time() * 1000)
+    ts = now - 86_400_000 * np.arange(n - 1, -1, -1)
+    def mk(slope):
+        c = np.linspace(100, 100 + slope * n, n)
+        return pd.DataFrame({"timestamp": ts, "open": c, "high": c, "low": c,
+                             "close": c, "volume": 1.0})
+    panel, cols = _panel({"BTC": mk(1.0), "AAA": mk(2.0)}, ["BTC", "AAA"])
+    assert len(panel) == n - 1          # barra in formazione scartata
+    assert int(panel["timestamp"].iloc[-1]) == int(ts[-2])
+
+
 def test_rotation_no_warn_on_cold_start(tmp_path, monkeypatch):
     calls = []
     monkeypatch.setattr("src.live.telegram_notifier.notify_event",