FIX: Have SeedCorpus._original_index track correct indices

[hanyuone]

When using SeedCorpus, we give it a set of seeds that corresponds 1-to-1 to a batched VerifiableModel's labeled_input (as in this example from ipynb/vnnlib_fuzzer.ipynb):

initial_seeds = [
    LabeledInputTensor(tensor=images[i:i+1], label=labels[i:i+1])
    for i in range(B)
]

However, SeedCorpus does deduplicating, and then stores self._original_index as the indices of the deduplicated list of seeds, instead of the original. For example, if we set the number of samples to 4, the images in labeled_input would correspond to images [0, 0, 0, 0, 1, 1, 1, 1, 2, 2, 2, 2, 3, 3, 3, 3]. However, _original_index = [0, 1, 2, 3] because of deduping.

This means that there is no longer a 1-to-1 correspondence between the indices in _original_index and the batched input spec, which means when we fuzz with BOX input specs, we use the wrong lb and ub. Input tensors now "blend together" when fuzzing:

To replicate this bug, run the following Python file as a Jupyter notebook, in the ipynb folder of ACT: https://gist.github.com/hanyuone/8a551837e90091ac720aa89c9f397c1c

This PR fixes the above bug by keeping track of original indices when deduplicating.

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.
✅ Project coverage is 73.51%. Comparing base (a5bf2e2) to head (e040b5b).

Additional details and impacted files

@@           Coverage Diff           @@
##             main     #105   +/-   ##
=======================================
  Coverage   73.51%   73.51%           
=======================================
  Files          91       91           
  Lines       20506    20508    +2     
=======================================
+ Hits        15075    15077    +2     
  Misses       5431     5431           

Flag	Coverage Δ
bab	45.71% <8.33%> (-0.01%)	⬇️
backend-float32	49.52% <8.33%> (-0.01%)	⬇️
backend-float64	49.56% <8.33%> (-0.01%)	⬇️
frontend	32.91% <100.00%> (+<0.01%)	⬆️
pipeline-fuzz	20.52% <100.00%> (+<0.01%)	⬆️
pipeline-verify	39.47% <8.33%> (+<0.01%)	⬆️

Flags with carried forward coverage won't be shown. Click here to find out more.

Files with missing lines	Coverage Δ
act/pipeline/fuzzing/corpus.py	87.96% <100.00%> (+0.22%)	⬆️

---

Continue to review full report in Codecov by Harness.

Legend - Click here to learn more
Δ = absolute <relative> (impact), ø = not affected, ? = missing data
Powered by Codecov. Last update a5bf2e2...e040b5b. Read the comment docs.

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.

[guanqin-123]

it is good to me.