v0.8.3

Merge pull request #6 from EffectiveEquivalent/claude/node-executable…

v0.8.2

Merge pull request #5 from EffectiveEquivalent/claude/node-executable…

v0.8.1

Merge pull request #4 from EffectiveEquivalent/claude/node-executable…

v0.7.0

v0.7.0

v0.6.3

v0.6.3: auto-fit spends its bands across the whole range

v0.5.2

v0.5.2: auto-fit uses the bands you ask for

v0.5.1

v0.5.1: no more stacked auto-fit bands

v0.5.0

v0.5.0: treble listening test, whole-target fit, spec-sheet layout

v0.4.0

v0.4.0: auto-fit rewrite

v0.3.0

v0.3.0: support for the second-generation cables

Adds a second wire protocol alongside the original one. The newer cables
speak HID report 0x4B over the interrupt endpoints rather than report 0x01
over the control pipe, carry 8 bands instead of 9, and have a real preamp
register, so none of the coefficient trickery gen-1 needs applies to them.

Verified on hardware:
  - Moondrop FreeDSP Mini  35D8:98D4  8 bands, +/-12 dB, PK/LS/HS, 5 voicings
  - CrinEar Protocol Micro 3302:C20F  8 bands, +/-10 dB, PK/LS/LPF/HPF

The generation is settled by asking the cable, not by matching a product
id, so other cables on the same chipset family should work unlisted.

Band count, gain range and filter types now follow the connected cable
rather than assuming gen-1's fixed 9 bands, and the app offers a choice
when more than one cable is attached.

Gen-2 writes are verified by reading the bank back, so the reported count
is bands the cable is actually holding rather than frames believed sent.

Also fixes a read correlation bug: replies are tagged only with the command
they answer, and reading a band and reading the active EQ slot share one,
so a stale reply would satisfy the next request and shift every answer
after it. The queue is drained before each request and the band index the
cable echoes back is now checked.

Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>