[SUPPORT] Datasource incremental subsequent read same as first read

hudi 0.12.2
spark 3.2.1
-----------------

Once an incremental read is made, all subsequent read on the table will remain the same. Likely something about incremental is cached.
NOTE: this behavior does not happens when loading the table by path (instead of with database/table)

```scala
// query 1
spark.read.format("hudi")
.option("hoodie.metadata.enable","true")
.table("database.hudi_table").count()
// 1000

// query 2
spark.read.format("hudi")
.option("hoodie.metadata.enable","true")
.option("hoodie.datasource.query.type","incremental")
.option("hoodie.datasource.read.begin.instanttime","20230203191804078")
.table("database.hudi_table").count()
// 200

// query 3
spark.read.format("hudi")
.option("hoodie.metadata.enable","true")
.table("database.hudi_table").count()
// 200 should be 1000

// query 4 after restarting the spark shell
spark.read.format("hudi")
.option("hoodie.metadata.enable","true")
.table("database.hudi_table").count()
// 1000
```

Also weird, the number of tasks is way smaller if query 2 is run in a fresh spark session.
- query 1 before query 2, then 25k tasks
- query 2 without query 1, then 17 tasks

Start by reproducing the four Spark reads in Hudi 0.12.2 with Spark 3.2.1, comparing database/table loading with path loading and recording the task counts in fresh and reused sessions. Done means a later non-incremental table read returns 1000 rather than the incremental result of 200, with the task-count difference understood and resolved.

Written by the indexing model from the issue text.