Fix plot_tree crash on truncated and multi-output SGTRegressor nodes

[upadhyan]

Fixes #82.

Root cause

_draw_leaf_text formatted node['value'] as one float. tree_export sets value=None on internal regressor nodes and a per-output list on multi-output leaves, and plot_tree draws nodes cut off by max_depth as leaves, so both cases raised TypeError. Classifiers didn't crash because class_counts is filled in on internal nodes too, but multi-output classifiers silently showed only output 0's class (and class_names=True produced garbled labels).

Changes

• Truncated regressor internals render as ….
• Multi-output leaves render one entry per output as [a, b], for regressors and classifiers. plot_tree resolves class names per output; pass one list per output for multi-output classifiers (same convention as export_text).
• The per-output and list formatting helpers move out of export_text, so both renderers share them.
• The plot_tree docstring states that n= is a rounded weighted count for classifiers and an unweighted count for regressors. Unifying the two needs a change to the C++ export, which is left out of this PR.

Why no test caught it

The only max_depth test used a classifier, and no plot_tree test fitted a 2-D y.

Tests

• New parametrized regressor test (1 or 2 outputs × max_depth None or 1) and a multi-output classifier test in tests/test_plot_tree.py. Both fail on main and pass on this branch.
• test-checker blind tests for _draw_leaf_text, promoted to tests/regression/: 9 fail on main and pass here, 30 pass on both, none fail here.
• Full suite: 1246 passed, 2 skipped.

🤖 Generated with Claude Code