[BUG] Crew.test crashes or misattributes scores when a ConditionalTask is skipped

Finding: In an external report, Crew.test() on crewAI main at commit 140bfcb (reported 2026-10-09; Windows 11, Python 3.13, uv venv) can raise IndexError ("list index out of range") after all iterations complete when a ConditionalTask is skipped in at least one iteration. Per the same report, scores for later tasks can land in the wrong rows, and a trailing skip in only one run can omit another run's trailing score from the crew average [https://github.com/crewAIInc/crewAI/issues/7997]. Reported reproduction shape: a crew with a regular task, a ConditionalTask whose condition sometimes returns False, and another regular task, then crew.test(2, eval_llm=...) so the conditional task is skipped in at least one iteration. The error is reported in CrewEvaluator.print_crew_evaluation_result() [https://github.com/crewAIInc/crewAI/issues/7997]. Author's hypothesis (not maintainer-confirmed): the evaluator attaches a callback to every task, but a skipped ConditionalTask never invokes it, so that run's score list is shorter. The printing code indexes the per-run score lists by every task position, which would explain both the IndexError and the shifted scores [https://github.com/crewAIInc/crewAI/issues/7997]. Expected behavior per the report: skipped tasks display '-', tasks that ran keep their scores, and per-task and crew averages use only existing scores [https://github.com/crewAIInc/crewAI/issues/7997]. Fix status: the author proposes storing scores at their task indices with placeholders for skipped tasks and computing averages from actual scores, in PR #7996. The author says regression tests using mocked model calls on the real Crew.test()/kickoff() path raised IndexError before the fix, that targeted evaluator tests pass afterward (7 passed), and that a separate trailing-skip case showed a too-low overall average before the fix [https://github.com/crewAIInc/crewAI/issues/7997]. These results are author-reported. The supplied sources do not show that the PR is merged or released, and there is no maintainer response. Limitations: this is a single external report. KFA did not reproduce it, and no independent reproduction exists. The author states the issue and PR were prepared with AI assistance. Affected released versions are unknown, and no workaround is reported. Comment coverage was partial.