rdrecord: allow smooth_frames=False for multi-segment records.

When reading a multi-segment, multi-frequency record, we want to have
the option of reading each signal at its original sampling frequency,
which requires using smooth_frames=False.  Previously this simply
wasn't allowed, either with or without multi-to-single conversion.

To do this, we need to ensure each segment is loaded in the
appropriate (smooth or non-smooth) mode (which formerly would have
failed if certain segments *didn't* contain multiple samples per
frame.)

After loading the segments, we must invoke multi_to_single, if
desired, in the appropriate mode.

Author: Benjamin Moody

wfdb/io/record.py

@@ -510,13 +510,6 @@ def check_read_inputs(
                 "return_res must be one of the following when physical is True: 64, 32, 16"
             )
 
-        # Cannot expand multiple samples/frame for multi-segment records
-        if isinstance(self, MultiRecord):
-            if not smooth_frames:
-                raise ValueError(
-                    "This package version cannot expand all samples when reading multi-segment records. Must enable frame smoothing."
-                )
-
     def _adjust_datetime(self, sampfrom):
         """
         Adjust date and time fields to reflect user input if possible.
@@ -4226,6 +4219,7 @@ def rdrecord(
                     channels=seg_channels[i],
                     physical=physical,
                     pn_dir=pn_dir,
+                    smooth_frames=smooth_frames,
                     return_res=return_res,
                 )
 
@@ -4242,7 +4236,9 @@ def rdrecord(
         # Convert object into a single segment Record object
         if m2s:
             record = record.multi_to_single(
-                physical=physical, return_res=return_res
+                physical=physical,
+                expanded=(not smooth_frames),
+                return_res=return_res,
             )
 
     # Perform dtype conversion if necessary