Improvements to ADPCM-B implementation

[aaronsgiles]

Specific fixes:

• Implement EOS and PLAYING flags as sticky bits that require manual clearing on all implementations.
• Implement correct end-of-sample behavior, skipping final 3 nibbles.
• Implement read access patterns that match hardware tests.
• Match behavior of "dummy" reads to hardware tests.
• Hook up verified edge case behaviors for over-reading and over-writing data.
• Add framework for running ADPCM tests to compare against hardware results.

[aaronsgiles]

Continuing conversation from https://github.com/aaronsgiles/ymfm/pull/38 over here.

[hyano]

I've added test for another unusual cases which READ & REC=1.

• https://github.com/hyano/opna-analyze/blob/main/test/mem_rw16.py
• https://github.com/hyano/opna-analyze/blob/main/log/mem_rw16_mden/mem_rw16.txt

In that condition, I observed:

• No memory access happens (confirmed by MDEN signal).
• The internal data buffer becomes empty after reading 2bytes and BRDY becomes 0.

It would be a hint for behavior of internal data buffer.

mem_rw16.cpp.txt