Post-open verification previously failed closed on any buffer-size delta on
every backend. The wedge this guards against (a driver accepting a request
without changing its buffer, then blocking the next in-place request) is
ASIO behaviour; ALSA rounds requests to period constraints and CoreAudio
can clamp, and both previously worked by storing the driver-adjusted
actuals. Keep exact buffer equality for ASIO only; other backends log and
accept the adjusted size. Rate and channel-mask verification stay strict
everywhere.
Also from review: print the real `options.compatible` in the live-probe log
instead of a hard-coded 1, and document that the live-endpoint reuse in
probeDual is safe only under runDeviceLifecycleOp's message-thread
serialisation (PR #113).
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Address PR #114 review:
- failClosed now stores duplexMode=false so a failed reconfigure cannot
leave the engine reporting duplex-active on a closed device.
- Drop the dead BigInteger initializers in the verify block.
- Move the applyDuplex locate-guard ahead of the source slice in the
lifecycle test so marker drift fails with a clear message.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
