Step 24E: isolate interpreter-only capture logic

lynglib/src/commonMain/kotlin/net/sergeych/lyng/BlockStatement.kt

@@ -33,6 +33,7 @@ class BlockStatement(
         if (captureSlots.isNotEmpty()) {
             val applyScope = scope as? ApplyScope
             for (capture in captureSlots) {
+                // Interpreter-only capture resolution; bytecode paths must use captureRecords instead.
                 val rec = if (applyScope != null) {
                     applyScope.resolveCaptureRecord(capture.name)
                         ?: applyScope.callScope.resolveCaptureRecord(capture.name)

lynglib/src/commonMain/kotlin/net/sergeych/lyng/Compiler.kt

@@ -7034,6 +7034,7 @@ class Compiler(
                     val captureBase = captureContext ?: closure
                     if (captureBase != null && captureSlots.isNotEmpty()) {
                         for (capture in captureSlots) {
+                            // Interpreter-only capture resolution; bytecode functions do not use resolveCaptureRecord.
                             val rec = captureBase.resolveCaptureRecord(capture.name)
                                 ?: captureBase.raiseSymbolNotFound("symbol ${capture.name} not found")
                             context.updateSlotFor(capture.name, rec)

lynglib/src/commonMain/kotlin/net/sergeych/lyng/Scope.kt

@@ -170,6 +170,9 @@ open class Scope(
     }
 
     internal fun resolveCaptureRecord(name: String): ObjRecord? {
+        if (captureRecords != null) {
+            raiseIllegalState("resolveCaptureRecord is interpreter-only; bytecode captures use captureRecords")
+        }
         return chainLookupIgnoreClosure(name, followClosure = true, caller = currentClassCtx)
     }