HDF5 Data reader enhancements

[bvanessen]

Added a check in the HDF5 data reader to check that the metadata for
each field actually matches the dimensions of the data fields.

Added a helper function for conduit to allow the calculation of a product of a data array's elements.

[benson31]

Suggested change

	int n_elts = node[pathname].dtype().number_of_elements();
	conduit::int64_array data_array_dims = metadata[HDF5_METADATA_KEY_DIMS].value();
	auto expected_n_elts = data_array_prod(data_array_dims);
	int const n_elts = node[pathname].dtype().number_of_elements();
	conduit::int64_array const data_array_dims = metadata[HDF5_METADATA_KEY_DIMS].value();
	auto const expected_n_elts = data_array_prod(data_array_dims);

[benson31]

Suggested change

	namespace conduit {
	template <typename T>
	T
	data_array_prod(DataArray<T> a)
	{
	T res = 1;
	for(index_t i = 0; i < a.number_of_elements(); i++)
	{
	const T &val = a.element(i);
	res *= val;
	}
	return res;
	}
	} // conduit
	namespace lbann {
	template <typename T>
	T
	data_array_prod(DataArray<T> const& a)
	{
	T res = 1;
	for(index_t i = 0; i < a.number_of_elements(); i++)
	{
	const T &val = a.element(i);
	res *= val;
	}
	return res;
	}
	} // namespace lbann

IMO we shouldn't be extending their namespace; just put it in lbann. Also, conduit::DataArray doesn't own any data, so it's just a "shallow copy" (pointer and size), but the "expected" API would probably be pass-by-const-reference. (I'm moderately surprised that they don't expose begin() and end() functions (and suitable iterators) for this container!)

[benson31]

Might be good to highlight in the output which is the observed vs the expected value -- shouldn't have to look in source code to decipher warnings...