## What changes were proposed in this pull request?

This fix tries to address the issue in PySpark where `spark.python.worker.memory`
could only be configured with a lower case unit (`k`, `m`, `g`, `t`). This fix
allows the upper case unit (`K`, `M`, `G`, `T`) to be used as well. This is to
conform to the JVM memory string as is specified in the documentation .

## How was this patch tested?

This fix adds additional test to cover the changes.

Author: Yong Tang <yong.tang.github@outlook.com>

Closes #12163 from yongtang/SPARK-14368.