Skip _metadata struct fields on data skipping

[uros7251brick]

Which Delta project/connector is this regarding?

• Spark
• Standalone
• Flink
• Kernel
• Other (fill in here)

Description

If we have a table with an integer column file_name and run a query like:

select *
from file_metadata_test
where _metadata.file_name = 'part-00000-fab85aae-5302-49c9-8e4d-747da36e5ae9.c000.zstd.parquet'

we get an error:

[[CAST_INVALID_INPUT](https://docs.databricks.com/error-messages/error-classes.html#cast_invalid_input)] The value 'part-00000-fab85aae-5302-49c9-8e4d-747da36e5ae9.c000.zstd.parquet' of the type "STRING" cannot be cast to "BIGINT" because it is malformed. Correct the value as per the syntax, or change its target type. Use `try_cast` to tolerate malformed input and return NULL instead. SQLSTATE: 22018

This happens because FileSourceStrategy#rebindFileSourceMetadataAttributesInFilters flattens _metadata struct, leaving just file_name, and then DataSkippingReader tries to use min-max stats of the actual file_name column to prune files and fails when it tries to cast the right-hand side of the equality predicate.

What changes were proposed in this pull request?

Fix the bug by not using dataFilters involving _metadata fields for data skipping.

How was this patch tested?

I added a new test case to DataSkippingDeltaTests.scala.

Does this PR introduce any user-facing changes?

No.

[chirag-s-db]

Nice fix, please also get a review from @tomvanbussel to make sure there aren't any potential DML implications

[tomvanbussel]

Nit: Why not just MetadataAttribute(_) if you're not intending to do anything with the name?

[uros7251brick]

you're right, i'll change it.