Pass storage_options to semantic dedup parquet grouping

[praateekmahajan]

In semantic dedup KMeans, break_parquet_partition_into_groups accepts storage_options, but KMeansReadFitWriteStage.process_batch and _fit_pass call it without passing self.input_storage_options.

For S3-compatible inputs, FilePartitioningStage and _read_group can use read_kwargs["storage_options"], but the grouping metadata read may fall back to ambient AWS config instead.

Fix: pass storage_options=self.input_storage_options into the two break_parquet_partition_into_groups(...) calls in semantic/kmeans.py.