Question about NaN GPU counts and End Times

**Question 1:**

In `pai_task_table`, I have noticed that 223965 of the 1261050 rows have NaN `plan_gpu` values.

Is there a reason for these values being marked as NaN? Is there any additional information that can be obtained from this?

It seems in `cluster-trace-gpu-v2020/analysis/analysis.ipynb` that these rows are dropped from analysis.

**Question 2:**

I noticed that 299948 *other* rows have NaN `end_time` values. This set of rows includes failed tasks.

For those failed tasks, what does a NaN `end_time` mean?

Start by reading cluster-trace-gpu-v2020/analysis/analysis.ipynb and inspecting the referenced pai_task_table fields and rows with NaN plan_gpu or end_time. Done means determining and documenting what each NaN value represents, including the failed-task cases, and explaining whether dropping those rows is appropriate for analysis.

Written by the indexing model from the issue text.