[dattri.algorithm] Adapt `vmap` to IFAttributor

[tingwl0122]

Description

This PR utilizes vmap to compute gradient within IF.

1. Motivation and Context

To speed up gradient computation in IF. (basically, borrow the idea from #72 )

2. Summary of the change

• modify test file correspondingly
• add an example for CIFAR-2 for IF
• let IF can compute grad using vmap
• remove projector functionality from IF (cannot be used with ihvp)
• remove the usage of train_gradient_batch_size and test_gradient_batch_size

3. What tests have been added/updated for the change?

• [x] Unit test: Typically, this should be included if you implemented a new function/fixed a bug.
• [x] Application test: If you wrote an example for the toolkit, this test should be added.

[tingwl0122]

Results of running example/cifar2_resnet9/influence_function_noisy_label.py: (also similar to random checking)

Peak memory usage: 39243.571712 MB Attribution time: 136301.74650335312

Checked Data Sample Found flipped Sample
0 0
100 9
200 25
300 35
400 40
500 46
600 56
700 65
800 76
900 90

[TheaperDeng]

Please fix the conflict. Then I think it should be good for now.