STM32 ring-buffer enhancement discussion

[elagil]

This is a follow-up to the discussion in #3336 by @liarokapisv

Feel free to add further people to the discussion.

Remaining issues/ideas with the current ring-buffer implementation include

• clear() on overrun causes issues with higher-level drivers that handle data in groups of samples. For example, a SAI driver may handle data in groups of two samples (left/right). The ring-buffer has no concept of grouping, so on clear() it will cause random read/write index offsets - in the case of the SAI example, causing left/right switching.
• Usefulness of write_immediate() is not clear, should it be removed?
• The ring buffer may pause automatically on overrun. The ring buffer shall then be explicitly restarted by a user when new data is available.