Apply data offset for list-of-pointers at access time rather than ListReader creation time.

Baking this offset into `ptr` reduced ops needed at access time but made the interpretation of `ptr` inconsistent depending on what type of list was expected.
diff --git a/c++/src/capnp/layout.c++ b/c++/src/capnp/layout.c++
index 7fa5b4e..fef6f48 100644
--- a/c++/src/capnp/layout.c++
+++ b/c++/src/capnp/layout.c++
@@ -2333,10 +2333,6 @@
             break;
 
           case ElementSize::POINTER:
-            // We expected a list of pointers but got a list of structs.  Assuming the first field
-            // in the struct is the pointer we were looking for, we want to munge the pointer to
-            // point at the first element's pointer section.
-            ptr += tag->structRef.dataSize.get();
             KJ_REQUIRE(tag->structRef.ptrCount.get() > ZERO * POINTERS,
                        "Expected a pointer list, but got a list of data-only structs.") {
               goto useDefault;
diff --git a/c++/src/capnp/layout.h b/c++/src/capnp/layout.h
index c8d533c..7a27f68 100644
--- a/c++/src/capnp/layout.h
+++ b/c++/src/capnp/layout.h
@@ -1227,8 +1227,12 @@
 }
 
 inline PointerReader ListReader::getPointerElement(ElementCount index) const {
+  // If the list elements have data sections we need to skip those. Note that for pointers to be
+  // present at all (which already must be true if we get here), then `structDataSize` must be a
+  // whole number of words, so we don't have to worry about unaligned reads here.
+  auto offset = structDataSize / BITS_PER_BYTE;
   return PointerReader(segment, capTable, reinterpret_cast<const WirePointer*>(
-      ptr + upgradeBound<uint64_t>(index) * step / BITS_PER_BYTE), nestingLimit);
+      ptr + offset + upgradeBound<uint64_t>(index) * step / BITS_PER_BYTE), nestingLimit);
 }
 
 // -------------------------------------------------------------------