fix: pin the waveform arbitration in a testable predicate, close round-4 review minors

Extracts resolveWaveformClaim (core/instrument/ui/spline_edit) so the shell's node/tab/marker click resolution is unit-tested directly, not just its input geometry; folds the staged-envelope node into it; fixes comment accuracy, a cost regression, and test fidelity issues.

src/shell/instrument/editor_session.cpp

@@ -186,8 +186,12 @@ ReaSamplerEditor::SetupMarkers ReaSamplerEditor::pickedMarkers(std::int64_t fram
     // Seed from the bank's intrinsic loop (fact about the file), then let the parameter set's
     // override win (the instrument's performance choice). Read the loop intrinsic from the
     // live bank blob (the same path selectSample uses); when that is not readable (extension
-    // absent / not yet parsed) the instance-owned ref carries the same intrinsics.
-    if (processor_) {
+    // absent / not yet parsed) the instance-owned ref carries the same intrinsics. Skipped
+    // entirely once an override is already set — it would just be overwritten below, and the
+    // bridge read + JSON parse it costs is real (mouseDownWaveform's arbitration calls this on
+    // every waveform click, not just marker grabs, to know whether a tab or marker candidate
+    // hits at all).
+    if (processor_ && !params_.loopOverride) {
         std::optional<SelectedSample> sel;
         auto banksJson =
             processor_->bridge().readReasamplerExtState(reasampler::kProjExtBanksKey);