Setting partitioned by (partition_path) with nonpartitioned keygenerator in spark-sql will cause the colum to be null

When creating the table, for example: 
{code:java}
create table hudi_cow_pt_tbl (
id bigint,
name string,
ts bigint,
dt string,
hh string
) using hudi
tblproperties (
type = 'cow',
primaryKey = 'id',
preCombineField = 'ts'
 hoodie.table.keygenerator.class = 'org.apache.hudi.keygen.NonpartitionedKeyGenerator'
)
partitioned by (dt) {code}
When attempting to cache the dataframe I read, I got the error 

 
{code:java}
assertion failed: Empty partition column value in 'partition_path='
java.lang.AssertionError: assertion failed: Empty partition column value in 'partition_path=' {code}
 

This will cause dt to always be null when you read the record. I don't know if the data is stored as null or just reads as null. If this is due to implementation issues and the only fix would be to fail the table creation, I think that is preferable to the current behavior.  

## JIRA info

- Link: https://issues.apache.org/jira/browse/HUDI-5263
- Type: Bug

Start by reproducing the provided Spark SQL table definition and dataframe cache/read sequence, checking whether the partition_path value is written or only reconstructed as null. Trace the interaction between the partitioned-by declaration and NonpartitionedKeyGenerator. Done means the invalid combination is rejected clearly or dt is preserved correctly when records are read.

Written by the indexing model from the issue text.