Rename `sum_hess` field to `weighted_data_count`

When building a treelite model, there is a method `data_count` which takes as input:
```
data_count – Number of data points
```

But if a given tree was fitted to data with sample weights, one would want it to instead represent the sum of weights among the observations, which will not necessarily be integral.

There is a `sum_hess` which takes float, but it's not exactly the same thing.

Search the repository for `sum_hess` and `data_count` to find their declarations, uses, and any related tests or documentation. Update the field naming consistently so the replacement represents a potentially non-integral sum of sample weights, then run the affected test suite and verify model building still works.

Written by the indexing model from the issue text.