[SPARK-20233][SQL] Apply star-join filter heuristics to dynamic programming join enumeration

## What changes were proposed in this pull request?

Implements star-join filter to reduce the search space for dynamic programming join enumeration. Consider the following join graph:

```
T1       D1 - T2 - T3
  \     /
    F1
     |
    D2

star-join: {F1, D1, D2}
non-star: {T1, T2, T3}
```
The following join combinations will be generated:
```
level 0: (F1), (D1), (D2), (T1), (T2), (T3)
level 1: {F1, D1}, {F1, D2}, {T2, T3}
level 2: {F1, D1, D2}
level 3: {F1, D1, D2, T1}, {F1, D1, D2, T2}
level 4: {F1, D1, D2, T1, T2}, {F1, D1, D2, T2, T3 }
level 6: {F1, D1, D2, T1, T2, T3}
```

## How was this patch tested?

New test suite ```StarJOinCostBasedReorderSuite.scala```.

Author: Ioana Delaney <ioanamdelaney@gmail.com>

Closes #17546 from ioana-delaney/starSchemaCBOv3.