convert_dtype: don't change p_signal when smooth_frames is false.

If physical is true, and smooth_frames is false, and e_p_signal is
None, this function should raise an exception.  It should not silently
modify p_signal instead.

The previous code didn't make any sense: p_signal is a numpy array, so
you can't change its data type by assigning to a slice of it, and even
if that conversion were possible, the slicing was being done along the
wrong axis.  This reverts a dubious change in commit
725e7a6, which seems unrelated to the
rest of that commit.

Author: Benjamin Moody

wfdb/io/_signal.py

@@ -705,14 +705,9 @@ def convert_dtype(self, physical, return_res, smooth_frames):
                 if current_dtype != return_dtype:
                     self.p_signal = self.p_signal.astype(return_dtype, copy=False)
             else:
-                if self.e_p_signal is not None:
-                    for ch in range(self.n_sig):
-                        if self.e_p_signal[ch].dtype != return_dtype:
-                            self.e_p_signal[ch] = self.e_p_signal[ch].astype(return_dtype, copy=False)
-                else:
-                    for ch in range(self.n_sig):
-                        if self.p_signal[ch].dtype != return_dtype:
-                            self.p_signal[ch] = self.p_signal[ch].astype(return_dtype, copy=False)
+                for ch in range(self.n_sig):
+                    if self.e_p_signal[ch].dtype != return_dtype:
+                        self.e_p_signal[ch] = self.e_p_signal[ch].astype(return_dtype, copy=False)
         else:
             return_dtype = 'int'+str(return_res)
             if smooth_frames: