chore: optimize dictionary access in strip_padding numpy (#3994)

* emplace field descriptors

* reserve sufficient capacity

* remove std::move

* properly iterate through dict

* make handle casting more explicit

* Revert to old dict api

include/pybind11/numpy.h

@@ -641,10 +641,14 @@ class dtype : public object {
             pybind11::str name;
             object format;
             pybind11::int_ offset;
+            field_descr(pybind11::str &&name, object &&format, pybind11::int_ &&offset)
+                : name{std::move(name)}, format{std::move(format)}, offset{std::move(offset)} {};
         };
+        auto field_dict = attr("fields").cast<dict>();
         std::vector<field_descr> field_descriptors;
+        field_descriptors.reserve(field_dict.size());
 
-        for (auto field : attr("fields").attr("items")()) {
+        for (auto field : field_dict.attr("items")()) {
             auto spec = field.cast<tuple>();
             auto name = spec[0].cast<pybind11::str>();
             auto spec_fo = spec[1].cast<tuple>();
@@ -653,8 +657,8 @@ class dtype : public object {
             if ((len(name) == 0u) && format.kind() == 'V') {
                 continue;
             }
-            field_descriptors.push_back(
-                {(pybind11::str) name, format.strip_padding(format.itemsize()), offset});
+            field_descriptors.emplace_back(
+                std::move(name), format.strip_padding(format.itemsize()), std::move(offset));
         }
 
         std::sort(field_descriptors.begin(),