H617A controls and more reliable Effect Studio operation

Added

• H617A Spectrum and Rolling fixed-colour controls.
• One-to-eight-colour palettes for seven newer music modes, Piano Keys Gradient,
  and Hopping background/no-colour controls.
• Seven named native DIY templates: Brilliant/Colorful, Colorful Sky, Meteor,
  Meteor Shower, Shine, Bloom DIY and Stack, preserving their native selectors.
• Segment colour-temperature service with fresh full-layout preservation checks.

Fixed

• Preserve mixed segment colours and relative brightness when putting back the
  previous appearance after a failed Effect Studio application. Capture fresh
  state first and verify supported readback, including master brightness.
• Clear incompatible retained Kelvin after mixed segment observations and
  reconcile colour writes that may have applied despite a transport error.
• Collect outstanding segment replies before subsequent controls, using
  response-driven completion and bounded deadlines rather than settling sleeps.
• Preserve newer Live previews when older Applies are saving or cancelled while
  waiting for control ownership.
• Keep complete known music bodies for replay and report uncertainty when
  original settings or full-body readback are unavailable.

Qualification

Tested through published HACS release candidates on H617A HW 3.01.01 / FW 3.02.24
with Home Assistant 2026.9.2: static/segment controls, all eleven music modes and
seven native-DIY API workflows, representative installed-browser controls,
overlapping Live edits, idle reconnect/polling and saved-effect restart persistence.
The owner confirmed representative music/noise response and Stack animation.
Final checks passed: 2,810 Python tests, 269 frontend unit tests, 33 browser tests
and 132 protocol replays (79 Python tests skipped).

Unknown physical IC geometry still disables dependent controls. New APK-derived
protocol coverage remains marked speculative; ACKs and selectors do not prove
complete resident-payload equality. Not every animation/control range was visually
qualified, and a 24-hour soak was not completed. Induced failure restoration remains
software-tested. Full evidence and limits: PR #301.