fix: remove zeros_t/ones_t/full_t — inconsistent with numpy naming

The _t suffix template functions introduced non-numpy names into the
public API. For dtype-flexible creation, use the native template functions
in core.h directly (zeros_like<T>, full<T>) with manual py::array pointer
extraction — this is the recommended path for all context builder migration.

Author: peng.li24

pycpp/core_py.h

@@ -54,7 +54,6 @@ py::array_t<T> empty_like(const py::array_t<T>& arr) {
 }
 
 /// numpy.zeros(shape, dtype=float, order='C', *, like=None)
-/// NOTE: convenient double default. For dtype flexibility, use zeros_t<T>(shape).
 inline py::array_t<double> zeros(const std::vector<py::ssize_t>& shape) {
     py::array_t<double> result(shape);
     zeros_like(static_cast<double*>(result.request().ptr), result.request().size);
@@ -75,29 +74,6 @@ inline py::array_t<double> full(const std::vector<py::ssize_t>& shape, double fi
     return result;
 }
 
-// Template counterparts — dtype-flexible creation for pybind11 modules
-// that need float32 or other dtypes. Bind as e.g. m.def("zeros_f32", &numpy::zeros_t<float>);
-template<typename T>
-py::array_t<T> zeros_t(const std::vector<py::ssize_t>& shape) {
-    py::array_t<T> result(shape);
-    zeros_like(static_cast<T*>(result.request().ptr), result.request().size);
-    return result;
-}
-
-template<typename T>
-py::array_t<T> ones_t(const std::vector<py::ssize_t>& shape) {
-    py::array_t<T> result(shape);
-    ones_like(static_cast<T*>(result.request().ptr), result.request().size);
-    return result;
-}
-
-template<typename T>
-py::array_t<T> full_t(const std::vector<py::ssize_t>& shape, T fill_value) {
-    py::array_t<T> result(shape);
-    numpy::full(static_cast<T*>(result.request().ptr), result.request().size, fill_value);
-    return result;
-}
-
 // Bool specializations
 // NOTE: _bool suffix — dtype-specific wrappers; pybind11 cannot deduce template
 // argument from a Python dtype keyword, so each dtype needs its own binding.