Phase 22: add non-destructive audio splitting, clip gain and fades, and cached waveforms

docs/ROADMAP.md

@@ -516,6 +516,23 @@ Provide focused, non-destructive audio editing tools for arranging and preparing
 - Clip boundaries do not produce unintended clicks from invalid offsets or discontinuities.
 - Editing and waveform generation never perform blocking work in the real-time audio callback.
 
+Implemented: transactional audio/silent-clip splitting with stable IDs and source
+offsets; persisted linear clip gain and explicit in/out fades in v8 (v1–v7 migrate
+to unity/no explicit fades); undoable sequencer menu/context/keyboard controls;
+immutable bounded source-peak summaries shared across clips and drawn from
+cached memory. Existing move/trim/duplicate/delete remain supported; very short
+clip hit testing now matches drawing. See AUDIO_FILES.md and PROJECT_FORMAT.md.
+Stretch/pitch processing and crossfades remain future work; waveform loading is
+synchronous on the control thread, and physical listening remains pending.
+
+Acceptance evidence:
+- [x] Core/GUI regressions cover split/duplicate/delete and snapshot undo.
+- [x] Exact byte comparison verifies source files survive all metadata edits.
+- [x] Boundary tests cover invalid offsets, short tails, one-frame media, explicit
+      fade endpoints, gain, block partitions, seeks and capacity rollback.
+- [x] Waveform construction stays on control-thread paths; redraw reads cached
+      peaks and audio rendering retains immutable/preloaded callback storage.
+
 ## Phase 23 — Modular Workflow and Routing Refinement
 
 ### Goal