audio: Refactor and improve the port-independent code

[jepler]

Some items that have come up repeatedly but have been addressed separately each time:

• Conversion of samples between signed & unsigned
• Conversion of mono samples to stereo
• Conversion between 8 and 16 bits (some ports even need other sizes, e.g., 10 bits (pico) or even non-power-of-two ranges (nrf))
• "re-blocking" audio to fit an underlying buffer size required by the API
• "ramping" to/from a quiescent value

We should more closely identify what we would like to do, find APIs to place (probably in common-hal/audiocore) that facilitates it, and then re-write each implementation to use the new APIs. This could also include modifying the existing APIs, e.g., how audiosamples work, to better fit our needs. This overall bug will probably require multiple PRs, so creating additional "sub-issues" is appropriate, especially when the specific sub-issue would be targeted to a release-based milestone.

[tannewt]

Some of this shared code will probably need to be platform aware too or at least weak. For example, the RP2040 has an interpolator that may be able to accelerate some of the conversion. Cortex-M4s also have some basic SIMD instructions that can be used to do multiple instructions at once. (I'm not sure if the compiler will automatically use them.)