[CH] Incorrect result when using df.queryExecution.executeCollect directly

### Backend

CH (ClickHouse)

### Bug description

```
def join(hint: String): DataFrame = {
sql(
s"""
|SELECT /*+ $hint */ *
|FROM testData t1
|LEFT OUTER JOIN
|testData2 t2
|ON key = a AND concat(value, b) = '12'
|""".stripMargin)
}

val df1 = join("SHUFFLE_HASH(t1)")
val res1 = df1.collect()
val res2 = df1.queryExecution.executedPlan.executeCollectPublic()
println(res1.size)
println(res2.size)
```

res1.size = 100, res2.size = 0

### Gluten version

_No response_

### Spark version

None

### Spark configurations

_No response_

### System information

_No response_

### Relevant logs

```bash

```

Start with the Scala reproducer in the issue and compare df1.collect() with df1.queryExecution.executedPlan.executeCollectPublic() for the ClickHouse backend. Investigate the relevant execution path and establish a test using the provided join and hint; done means both collection paths return the same result size.

Written by the indexing model from the issue text.