Matrix-free evaluation: Look over loops over vector components

[kronbichler]

As suggested by #16003 (comment) after a remark by @masterleinad we should look over loops of the kind

dealii/include/deal.II/matrix_free/fe_evaluation.h

Lines 3406 to 3420 in 4e1c906

	if (n_components == 1 || this->n_fe_components == 1)
	for (unsigned int i = 0; i < dofs_per_component;
	++i, dof_indices += n_lanes)
	for (unsigned int comp = 0; comp < n_components; ++comp)
	operation.process_dof_gather(dof_indices,
	*src[comp],
	0,
	values_dofs[comp][i],
	vector_selector);
	else
	for (unsigned int comp = 0; comp < n_components; ++comp)
	for (unsigned int i = 0; i < dofs_per_component;
	++i, dof_indices += n_lanes)
	operation.process_dof_gather(
	dof_indices, *src[0], 0, values_dofs[comp][i], vector_selector);

where we have two flavors, depending on whether we build multi-component problems from multiple vectors with a scalar FE (first case) or with a single vector from an FESystem (second case). We should be able to simplify the code in various places if we can use a single loop.