TabularSHAP hangs indefinitely

@memoryz is already working on this.

Since May 1, 2023.

Hi SynapseML team,

I am using TabularSHAP to explain predictions of the LightGBM classification model using SynapseML version `com.microsoft.azure:synapseml_2.12:0.9.5-35-e962330b-SNAPSHOT`

Based on the [tutorial](https://microsoft.github.io/SynapseML/docs/next/features/responsible_ai/Interpretability%20-%20Tabular%20SHAP%20explainer), I have prepared my code below. I am using only one executor with five cores to ensure the code doesn't fail due to concurrency, and I repartition all data frames to use a single partition.

```
from pyspark.ml import Pipeline
from pyspark.ml.feature import VectorAssembler
from synapse.ml.lightgbm import JavaMLReadable, LightGBMClassifier, LightGBMClassificationModel
from synapse.ml.explainers import TabularSHAP

model = LightGBMClassifier.load('')
train_data = spark.read.parquet('').repartition(1).cache() # type: about 15K records
features = [...list of features...]

vector_assembler = VectorAssembler(inputCols = features, outputCol = 'features').setHandleInvalid("skip")
pipeline = Pipeline(stages=[vector_assembler, model])
model = pipeline.fit(train_data)

explain_instances = model.transform(train_data.limit(5).repartition(1).cache()).cache()
train_sample = train_data.orderBy(F.rand()).limit(100).repartition(1).cache()

shap = TabularSHAP(
inputCols=features,
outputCol="shapValues",
numSamples=10,
model=model,
targetCol="probability",
targetClasses=[1],
backgroundData=train_sample
)

explained = shap.transform(explain_instances).cache()
```

On the last line, when I call `shap.transform()`, the notebook shows me the progress bar but waits forever. I think it gets into some lock, but I am not sure. But I am sure it doesn't wait for the data frames to be calculated, because, while I have removed these lines from the code, I call `.show()` to force caching...

What am I missing?

Thank you.