Update inference method involving prune tree prediction.

[zhi-bao]

What does this PR do?

This PR does the following two things:

1. Reimplements the inference method in LibMultiLabel. In the inference method of LibMultiLabel, the algorithm performs a single matrix multiplication between the batch instances and a weight matrix containing all classifiers. However, when the beam width is smaller than the root's degree, unnecessary computations occur on root child nodes that are unreachable. To address this, we reimplement the method by pruning those unreachable child nodes during computation with the given batch instances, improving efficiency.

2. Modifies the structure of the Tree Model based on the new implementation. In the tree model structure of LibMultiLabel, all nodes' classifiers needed to be flattened into a single weight matrix. LibMultiLabel uses hstack to flatten weights, which requires twice the memory due to duplication. However, in the new implementation, we only need to flatten the weights in the root's subtree, which requires less memory than flattening all nodes' classifiers, thereby reducing overall memory usage.

Test CLI & API (bash tests/autotest.sh)

Test APIs used by main.py.

• Test Pass
  • (Copy and paste the last outputted line here.)
• Not Applicable (i.e., the PR does not include API changes.)

Check API Document

If any new APIs are added, please check if the description of the APIs is added to API document.

• API document is updated (linear, nn)
• Not Applicable (i.e., the PR does not include API changes.)

Test quickstart & API (bash tests/docs/test_changed_document.sh)

If any APIs in quickstarts or tutorials are modified, please run this test to check if the current examples can run correctly after the modified APIs are released.