Handle missing values in dataframe with category dtype. (#7331)

* Replace -1 in pandas initializer.
* Unify `IsValid` functor.
* Mimic pandas data handling in cuDF glue code.
* Check invalid categories.
* Fix DDM sketching.

src/tree/updater_gpu_hist.cu

@@ -580,6 +580,7 @@ struct GPUHistMakerDevice {
       CHECK_LT(candidate.split.fvalue, std::numeric_limits<bst_cat_t>::max())
           << "Categorical feature value too large.";
       auto cat = common::AsCat(candidate.split.fvalue);
+      common::CheckCat(cat);
       std::vector<uint32_t> split_cats(LBitField32::ComputeStorageSize(std::max(cat+1, 1)), 0);
       LBitField32 cats_bits(split_cats);
       cats_bits.Set(cat);