Poking away at some ideas for #599.

Some thoughts from writing this:

• Managing the recursive child/dictionary lifecycle is probably best left to some "root holder" like the UniqueSchema. I tried making a Schema class but it's very hard to write things like Child() without some sharedness or a lot of copying. I went with ViewSchema() for argument's sake.
• For the SchemaBuilder I went with an approach that probably copies too much (but in theory lets you std::move() to avoid issues, sort of like std::vector will let you copy or move.

Example:

TEST(SchemaHpp, SchemaHppDump) {
  std::stringstream ss;

  auto schema = nanoarrow::schema::struct_({{"foofy", NANOARROW_TYPE_INT32}});
  nanoarrow::ViewSchema view(schema);

  ss << "metadata size: " << view.metadata().size() << std::endl;

  for (const auto& item : view.metadata()) {
    ss << item.first << ", " << item.second << std::endl;
  }

  for (const auto& child : view.children()) {
    ss << "child: " << child.name() << std::endl;
  }

  if (view.dictionary()) {
    ss << "has dictionary\n";
  }
}