Better API for specifying return data of signals

The current rdrecord API is rather confusing when dealing with more complex records/signals. Namely: multi-frequency signals.

We should create a better API to allow users to:

• Specify/know the exact return types/dimensions of the signals being read
• Deal with single and multi-frequency records consistently.

In addition, the current API is also limiting when reading single-frequency records, as there is no way for users to explicitly specify whether they want 1d or 2d numpy arrays.

We current have: https://github.com/MIT-LCP/wfdb-python/blob/0d42dfb4b2946625f00cbf500d830d374a201153/wfdb/io/record.py#L1715 where the only parameter that influences the return type is: smooth_frames

We've chatted about the issue here, where I suggested a new set of params:
https://github.com/MIT-LCP/wfdb-python/pull/313#issuecomment-1058363761

What do people think of the suggestion of having two params: return_dims and smooth_frames? return_dims will apply for both single and multi-frequency signals, and smooth_frames will apply only to multi-frequency signals. Given the combination of the params, the return value of the signals will be:

(1, True): List of 1d arrays. Guaranteed to be same length.
(1, False): List of 1d arrays. Not guaranteed to be same length for multi-frequency signals.
(2, True): Single 2d array.
(2, False): Illegal combination.

Read wfdb/io/record.py around line 1715 and the linked pull request discussion first, since the proposed parameters and compatibility questions are still under discussion. Done means establishing and documenting the return shapes for single- and multi-frequency signals across return_dims and smooth_frames, with the illegal combination handled consistently.