Use Span in gpu coordinate. (#4029)

* Use Span in gpu coordinate.

* Use Span in device code.
* Fix shard size calculation.
  - Use lower_bound instead of upper_bound.
* Check empty devices.

src/common/span.h

@@ -621,8 +621,8 @@ XGBOOST_DEVICE auto as_writable_bytes(Span<T, E> s) __span_noexcept ->  // NOLIN
   return {reinterpret_cast<byte*>(s.data()), s.size_bytes()};
 }
 
-}  // namespace common
-}  // namespace xgboost
+}  // namespace common NOLINT
+}  // namespace xgboost NOLINT
 
 #if defined(_MSC_VER) &&_MSC_VER < 1910
 #undef constexpr