### What changes were proposed in this pull request?
In https://issues.apache.org/jira/browse/SPARK-29283, we only show the error message of root cause to end-users through JDBC client. In some cases, it erases the straightaway messages that we intentionally make to help them for better understanding.
The root cause is somehow obscure for JDBC end-users who only writing SQL queries.
e.g
```
Error running query: org.apache.spark.sql.AnalysisException: The second argument of 'date_sub' function needs to be an integer.;
```
is better than just
```
Caused by: java.lang.NumberFormatException: invalid input syntax for type numeric: 1.2
```
We should do as Hive does in https://issues.apache.org/jira/browse/HIVE-14368
In general, this PR partially reverts SPARK-29283, ports HIVE-14368, and improves test coverage
### Why are the changes needed?
1. Do the same as Hive 2.3 and later for getting an error message in ThriftCLIService.GetOperationStatus
2. The root cause is somehow obscure for JDBC end-users who only writing SQL queries.
3. Consistency with `spark-sql` script
### Does this PR introduce _any_ user-facing change?
Yes, when running queries using thrift server and an error occurs, you will get the full stack traces instead of only the message of the root cause
### How was this patch tested?
add unit test
Closes#28963 from yaooqinn/SPARK-32145.
Authored-by: Kent Yao <yaooqinn@hotmail.com>
Signed-off-by: Wenchen Fan <wenchen@databricks.com>
59a70879c0