trees: Support regression with RandomForest

[jonnor]

The underlying C library supports this, but it need to wire it up correctly in the MicroPython modules. And need to chec that it saves and loads correctly into the CSV format. Should add a basic test/example as well.

[jonnor]

Some applications may be interested in having multi-output regression. For example for estimating X, Y coordinates.
Multiple-output is not supported directly, so one needs to use one-model-per-output. Prediction quality should be just as good - but it will require a little bit more compute/model size. However, it should be negligible in most cases.

It would be useful to show example of how to do that, both on training and inference side.
On the training side, there is a standard multioutput wrapper in scikit-learn. Example: https://scikit-learn.org/stable/auto_examples/ensemble/plot_random_forest_regression_multioutput.html

One would have a similar wrapper on the MicroPython side, to make it convenient.

[jonnor]

One example application of multi-output regression with Random Forest is indoor positioning using signal strength measurements. BLE and/or WiFi. Assuming that they are labeled with a location as continious parameters in 2d (or more). Some datasets only have a categorical location.

Example datasets for indoor localization:

• https://zenodo.org/records/10862916
• https://github.com/philotuxo/Position-Annotated-BLE-RSSI-Dataset
• https://data.nist.gov/od/id/mds2-2827
• https://zenodo.org/records/13317046

Example datasets for gas concentration:

• https://archive.ics.uci.edu/dataset/361/twin+gas+sensor+arrays
• https://archive.ics.uci.edu/dataset/322/gas+sensor+array+under+dynamic+gas+mixtures