fix: batched upload permanently truncates in-memory test run

[aerosta]

Summary

Large test runs are uploaded in batches, but post_test_run() mutates the caller's in-memory TestRun while doing so. It replaces test_run.test_cases or test_run.conversational_test_cases with only the first batch and clears test_run.prompts, but never restores that state afterward.

This means any code that reads the same TestRun object after upload can see a partial run instead of the full evaluated result.

This change snapshots the original state before batching and restores it in a finally block, so the full in-memory run is preserved whether upload succeeds or fails.

Changes

• deepeval/test_run/test_run.py: preserve and restore the original run state during batched upload
• tests/test_core/test_run/test_run_manager.py: add regression coverage for successful batched upload and mid-upload failure

Test plan

DEEPEVAL_TELEMETRY_OPT_OUT=1 python -m pytest tests/test_core/test_run/test_run_manager.py -q --tb=short

[vercel]

@aerosta is attempting to deploy a commit to the Confident AI Team on Vercel.

A member of the Team first needs to authorize it.