[FEA] Add diagnostic reason for GPU slowness in Profiler tool

[cindyyuanjiang]

We want to add support for top 1-3 use cases for diagnostic reason for GPU slowness in Profiler tool.

1. Highest stage time (descending order) contributors -- like spill amount per stage, skew (input /output bytes), shuffle bytes depending on the stage plan, GPU semaphore time, and number of tasks.
2. SQL Plan buffer time, operator time, number of input and output rows for execs in that stage.
3. Table sizes (row and column counts) for interesting operators in these long running stages (joins, aggregations etc).

Subtasks

• Create a stage-level summary view with memory/disk spilled, input/output bytes, shuffle read/write and gpu semaphore info (descending order of stage duration)
• Create an IO View which highlights CPU and GPU specific SQL/Operator metrics for Scans and Shuffles
• Create Filtered Operator and SQL Metric View for top 7 stages (highest duration)

[kuhushukla]

Can we break this down into subtasks?

On a second thought, we should combine the two tables shown as an example here. My original intent was to keep the first view simple but the latter table is not too bad for that