Fixes UnboundLocalError: local variable 'cv_pipeline' referenced before assignment when error in automl search

[angela97lin]

Closes #995

Before this PR, in the unlikely chance that _compute_cv_scores errors before setting cv_pipeline = pipeline.clone(), then we get an UnboundLocalError: local variable 'cv_pipeline' referenced before assignment in the later check if isinstance(cv_pipeline, BinaryClassificationPipeline) and cv_pipeline.threshold is not None. This PR addresses this by initializing it to None first before the try/except block.
`

[codecov]

Codecov Report

Merging #996 into main will increase coverage by 0.19%.
The diff coverage is 100.00%.

@@            Coverage Diff             @@
##             main     #996      +/-   ##
==========================================
+ Coverage   99.67%   99.86%   +0.19%     
==========================================
  Files         179      179              
  Lines        9424     9436      +12     
==========================================
+ Hits         9393     9423      +30     
+ Misses         31       13      -18     

Impacted Files	Coverage Δ
...alml/pipelines/prediction_explanations/__init__.py	100.00% <ø> (ø)
evalml/automl/automl_search.py	99.55% <100.00%> (+0.45%)	⬆️
evalml/tests/automl_tests/test_automl.py	100.00% <100.00%> (ø)
evalml/tests/component_tests/test_components.py	99.57% <0.00%> (+0.42%)	⬆️
.../automl_tests/test_automl_search_classification.py	100.00% <0.00%> (+0.45%)	⬆️
evalml/tests/pipeline_tests/test_pipelines.py	100.00% <0.00%> (+0.82%)	⬆️
...ests/automl_tests/test_automl_search_regression.py	100.00% <0.00%> (+1.06%)	⬆️
evalml/utils/gen_utils.py	100.00% <0.00%> (+2.53%)	⬆️
... and 2 more

---

Continue to review full report at Codecov.

Legend - Click here to learn more
Δ = absolute <relative> (impact), ø = not affected, ? = missing data
Powered by Codecov. Last update 93e7ebe...f36609e. Read the comment docs.

[freddyaboulton]

@angela97lin Looks good to me! I left a comment about how to write a unit test for this. Apart from that, this made it clear to me that we don't have a good way of catching errors that come from splitting the data. That might be intended but if not we should discuss what to do about that.

[angela97lin]

@freddyaboulton Agreed, we don't have a clear way of telling the user that the error stemmed from splitting data, but with this in place, at least we'll default to our catching / logging of pipeline errors :D

[dsherry]

@angela97lin great reproducer on the issue, and your explanation of the problem on here was helpful! This rocks. I left one comment about checking for nan scores in the test.

[jeremyliweishih]

LGTM -agree with Dylan's comment on checking for nan