Numba riddance

adresses #425 (closed)

• removes numba dependency/usage from saqc
• adds replacement mechanics to make up for performance loss
• see #425 (closed) for an overview over the loss-trade offs (witch are fairly minimal to non-existant in focus use-cases)

saqc/funcs/curvefit.py

@@ -176,17 +176,13 @@ def _fitPolynomial(
         fitted = to_fit.rolling(
             pd.Timedelta(window), closed="both", min_periods=min_periods, center=True
         ).apply(polyRollerIrregular, args=(centers, order))
-    else:
+    else: # if regular
         if isinstance(window, str):
             window = pd.Timedelta(window) // regular
         if window % 2 == 0:
             window = int(window - 1)
         if min_periods is None:
             min_periods = window
-        if len(to_fit) < 200000:
-            numba = False
-        else:
-            numba = True
 
         val_range = np.arange(0, window)
         center_index = window // 2
@@ -202,43 +198,21 @@ def _fitPolynomial(
             miss_marker = np.floor(miss_marker - 1)
             na_mask = to_fit.isna()
             to_fit[na_mask] = miss_marker
-            if numba:
-                fitted = to_fit.rolling(window).apply(
-                    polyRollerNumba,
-                    args=(miss_marker, val_range, center_index, order),
-                    raw=True,
-                    engine="numba",
-                    engine_kwargs={"no_python": True},
-                )
-                # due to a tiny bug - rolling with center=True doesnt work
-                # when using numba engine.
-                fitted = fitted.shift(-int(center_index))
-            else:
-                fitted = to_fit.rolling(window, center=True).apply(
-                    polyRoller,
-                    args=(miss_marker, val_range, center_index, order),
-                    raw=True,
-                )
+
+
+            fitted = to_fit.rolling(window, center=True).apply(
+                polyRoller,
+                args=(miss_marker, val_range, center_index, order),
+                raw=True,
+            )
             fitted[na_mask] = np.nan
         else:
             # we only fit fully populated intervals:
-            if numba:
-                fitted = to_fit.rolling(window).apply(
-                    polyRollerNoMissingNumba,
-                    args=(val_range, center_index, order),
-                    engine="numba",
-                    engine_kwargs={"no_python": True},
-                    raw=True,
-                )
-                # due to a tiny bug - rolling with center=True doesnt work
-                # when using numba engine.
-                fitted = fitted.shift(-int(center_index))
-            else:
-                fitted = to_fit.rolling(window, center=True).apply(
-                    polyRollerNoMissing,
-                    args=(val_range, center_index, order),
-                    raw=True,
-                )
+            fitted = to_fit.rolling(window, center=True).apply(
+                polyRollerNoMissing,
+                args=(val_range, center_index, order),
+                raw=True,
+            )
 
     data[field] = fitted
     worst = flags[field].rolling(window, center=True, min_periods=min_periods).max()