Roadmap: feature requests

This issue is aimed at keeping track of requested features.

Note to developers and contributors: If you plan to actively work to implement a requested feature, re-open the linked issue. **Everyone is welcome to work on any of the issues below.** For specifics, read the linked issues.

Note to maintainers: All issues with feature requests should be consolidated to this document. Close all new issues with feature requests and create corresponding new entries in the following checklist. (Don't forget to create a link to the closed issue.) This is so that the number of open issues is kept manageable. **Also make sure to attach `feature-request` label when closing feature requests**

- [ ] Multiple output regression (#2087)
- [ ] Rust binding (#3351)
- [ ] Lua binding (#1444)
- [ ] Save data counts inside XGBoost model (#3419)
- [x] Feature interaction constraint (#3135)
- [ ] Clarify behavior for trivial splits (#2914)
- [ ] Stratified distributed DMatrix (#2758)
- [ ] Dynamic `min_child_weight` parameter (#2714)

### Distributed learning
- [ ] Write a tutorial for distributed learning using Docker containerization (#3593)
- [ ] Provide a tutorial for using XGBoost with MPI (#3596)

### Python specific
- [x] Preserve `feature_names` when slicing DMatrix with `slice()` (#3124)
- [x] Pre-built binary wheels for Windows
- [x] Automatically set `ntree_limit = best_ntree_limit` for predictions in scikit-learn API (#3053)
- [x] Make prediction at each stage for a given data matrix (#2175)
- [x] Enable prediction on CPU-only machines with models trained with GPUs (#3342)
- [x] Booster.load_model() should not require buffer writeability (#3013)
- [ ] Interoperability with `sklearn.tree.export_graphviz` (#2981)

### JVM specific
- [x] Update documentation
- [x] Single-instance prediction interface (#3153)
- [x] Attach group info for each row in data frame (#3097)
- [x] Evaluate an arbitrary number of evaluation sets when training in XGBoost-Spark (#3231)
- [x] Pass additional parameters to XGBoostEstimator (#3202)
- [x] Do not expose `booster` parameter to XGBoost-Spark (#3209)
- [x] resolve external cache file for windows (#3038)
- [x] Float.NaN as missing value causes program crash (#3370)
- [x] add early stopping in XGBClassifier/Regressor (#2710)
- [ ] Do train/test-split by column-values in XGBoost4J-Spark (#3607)

### Documentation
- [x] Modernize XGBoost docs (#3444)
- [x] Document CSC/CSR format (#3331)
- [x] Document test system for developers (#3283)
- [x] Update the R tutorial "Discover your data" to reflect the deletion of RealCover (#3239)
- [x] Document usage of C/C++ API