2.10.2

Features:

• #417 and #579 Added separate actions to check saved connections, find all synths, check one synth's saved route, or search only for that synth. This makes it much quicker to reconnect a DIN synth powered on after KnobKraft and keeps existing MIDI input listeners intact during searches. Thanks to @RadekPilich for the request! The automated connection and cancellation tests pass; physical reconnection during an active scan still needs hardware verification.
• #478 and #578 Python adaptations can now provide named, preformatted patch views for the current-patch sidebar and patch comparison. Formatting is preserved, matching views are compared with JUCE's text diff, and the Korg MS2000 adaptation includes an unescaped payload view with offsets. Thanks to @RadekPilich for the idea and detailed examples!
• #468 and #580 Fixed the secondary MIDI output so program changes, bank select, notes, SysEx and realtime messages are forwarded independently of the MIDI log filter. Adaptation API level 3 also adds optional onPatchSelected and onPatchSent hooks, allowing a personalized adaptation to send controller setup messages when browsing or successfully sending a patch. Thanks to @RadekPilich for the original request, testing and Faderfox EC4 use case!

Bug fixes:

• #547 and #563 Corrected Korg R3 program and edit-buffer dump sizes. Valid hardware dumps are now recognized instead of leaving bank downloads at zero, while incomplete and malformed messages remain rejected. Thanks to @molenick for the report, the working implementation reference and the captured response! Automated protocol fixtures pass; a complete hardware capture is still welcome for the separate A1-request/B1-response observation.
• #582 and #583 Fixed a missing pybind11 STL converter that made Python adaptation bankSelect calls fail on macOS, as reported with the Sequential Prophet 12. Thanks to @kisielk for the report! The regression passes on macOS and Windows.
• #553 Updated JUCE from 8.0.12 to 8.0.15, including its Linux ALSA fix that falls back to legacy byte-stream MIDI when a destination does not support UMP. This addresses the all-message output failure isolated with the Kawai K5000S loopback test. Thanks to @firstnevyn for the logs, captures and careful testing! A K5000 hardware retest is still required; the separate Bank D import-before-detection and unsupported Multi-classification problems remain open.
• Improved macOS release packaging by retrying forced detach of temporary disk images and preserving the failure result when every attempt fails.