feat: TA-Lib candlestick patterns — gap-three-methods/stalled/stick-sandwich/takuri (part 8 of 9) (#140)

Adds five TA-Lib candlestick patterns, each a streaming `Indicator<Input = Candle, Output = f64>` emitting the family's uniform `±1.0 / 0.0` sign convention, fully wired across the Rust core, Python / Node / WASM bindings, fuzz target and reference tests.

- **Upside Gap Three Methods** (`CDLXSIDEGAP3METHODS`) — a 3-bar bullish continuation: two white candles gap up, then a black candle opens within the second body and closes within the first; bullish +1.
- **Downside Gap Three Methods** (`CDLXSIDEGAP3METHODS`) — the bearish mirror: two black candles gap down, then a white candle opens within the second body and closes within the first; bearish -1.
- **Stalled Pattern** (`CDLSTALLEDPATTERN`) — a 3-bar bearish reversal warning: two long white candles then a small white candle riding the shoulder, signalling the rally is stalling; bearish -1.
- **Stick Sandwich** (`CDLSTICKSANDWICH`) — a 3-bar bullish reversal: two black candles closing at the same level sandwich a white candle, marking a support floor; bullish +1.
- **Takuri** (`CDLTAKURI`) — a single-bar bullish reversal, a strict Dragonfly Doji with a negligible upper shadow and very long lower shadow; bullish +1.

Body and shadow thresholds follow the geometric house style (fixed fractions of the bar range) rather than TA-Lib's rolling averages. Upside / Downside Gap Three Methods share the `CDLXSIDEGAP3METHODS` code, so the second carries a manual CHANGELOG entry (as with Rising / Falling Three Methods).

Counter 279 → 284 (mod-count == lib counted block; FAMILIES total 274 → 279).

Stacked on #139 (`feat/cdl-lines`); base retargets to `main` once the predecessor merges.

bindings/python/tests/test_new_indicators.py

@@ -663,6 +663,26 @@ CANDLE_SCALAR = {
         lambda: ta.FallingThreeMethods(),
         lambda ind, h, l, c, v: ind.batch(c, h, l, c),
     ),
+    "UpsideGapThreeMethods": (
+        lambda: ta.UpsideGapThreeMethods(),
+        lambda ind, h, l, c, v: ind.batch(c, h, l, c),
+    ),
+    "DownsideGapThreeMethods": (
+        lambda: ta.DownsideGapThreeMethods(),
+        lambda ind, h, l, c, v: ind.batch(c, h, l, c),
+    ),
+    "StalledPattern": (
+        lambda: ta.StalledPattern(),
+        lambda ind, h, l, c, v: ind.batch(c, h, l, c),
+    ),
+    "StickSandwich": (
+        lambda: ta.StickSandwich(),
+        lambda ind, h, l, c, v: ind.batch(c, h, l, c),
+    ),
+    "Takuri": (
+        lambda: ta.Takuri(),
+        lambda ind, h, l, c, v: ind.batch(c, h, l, c),
+    ),
 }
 
 
@@ -1951,6 +1971,39 @@ def test_falling_three_methods_reference():
     assert t.update((12.0, 13.1, 11.9, 13.0, 1.0, 3)) == pytest.approx(0.0)
     assert t.update((12.5, 12.6, 8.9, 9.0, 1.0, 4)) == pytest.approx(-1.0)
 
+
+def test_upside_gap_three_methods_reference():
+    t = ta.UpsideGapThreeMethods()
+    assert t.update((10.0, 11.2, 9.8, 11.0, 1.0, 0)) == pytest.approx(0.0)
+    assert t.update((12.0, 13.2, 11.9, 13.0, 1.0, 1)) == pytest.approx(0.0)
+    assert t.update((12.5, 12.6, 10.4, 10.5, 1.0, 2)) == pytest.approx(1.0)
+
+
+def test_downside_gap_three_methods_reference():
+    t = ta.DownsideGapThreeMethods()
+    assert t.update((13.0, 13.2, 11.8, 12.0, 1.0, 0)) == pytest.approx(0.0)
+    assert t.update((11.0, 11.1, 9.8, 10.0, 1.0, 1)) == pytest.approx(0.0)
+    assert t.update((10.5, 12.6, 10.4, 12.5, 1.0, 2)) == pytest.approx(-1.0)
+
+
+def test_stalled_pattern_reference():
+    t = ta.StalledPattern()
+    assert t.update((10.0, 12.05, 9.9, 12.0, 1.0, 0)) == pytest.approx(0.0)
+    assert t.update((11.0, 14.05, 10.9, 14.0, 1.0, 1)) == pytest.approx(0.0)
+    assert t.update((14.0, 14.6, 13.95, 14.15, 1.0, 2)) == pytest.approx(-1.0)
+
+
+def test_stick_sandwich_reference():
+    t = ta.StickSandwich()
+    assert t.update((12.0, 12.1, 9.9, 10.0, 1.0, 0)) == pytest.approx(0.0)
+    assert t.update((10.5, 11.6, 10.4, 11.5, 1.0, 1)) == pytest.approx(0.0)
+    assert t.update((11.5, 11.6, 9.9, 10.0, 1.0, 2)) == pytest.approx(1.0)
+
+
+def test_takuri_reference():
+    t = ta.Takuri()
+    assert t.update((10.0, 10.05, 7.0, 10.0, 1.0, 0)) == pytest.approx(1.0)
+
 # --- Lifecycle ------------------------------------------------------------