Implementation of stft() function as stft_array and tfr_stft

Describe the new feature or enhancement

Currently stft operates on arrays, but is not the same API as wavelet, stockwell and multitaper. In addition, it does not have a tfr_stft analog , which operates on mne-python data structures.

The proposed PR would refactor stft to be named stft_array and have similar API to the tfr_array functions, and also implement an analogous tfr_stft function.

See discourse discussion with @drammock: https://mne.discourse.group/t/is-there-a-way-to-obtain-the-time-points-from-time-frequency-stft/2682

The stft is a nice function since it is super fast, and can operate on Raw object data within seconds.

Describe your proposed implementation

Tbd

Additional comments

related: https://github.com/mne-tools/mne-python/pull/8781. If this is accepted, then one can use tfr_stft to compute spectrograms of the data for certain frequency bands (e.g. gamma power shown as channels X time over a whole Raw dataset).

Start by reviewing the existing stft implementation and the tfr_array APIs, then compare the related pull request #8781 and the linked Discourse discussion. Define the scope for renaming stft to stft_array and adding tfr_stft for MNE-Python data structures; done means both APIs are implemented consistently and the requested Raw-data spectrogram use case is supported.