Remove usage of MicroPartition.cast_to_schema in read function

src/daft-micropartition/src/micropartition.rs

@@ -404,7 +404,7 @@ impl MicroPartition {
                     .map(|cols| cols.iter().map(|s| s.as_str()).collect::<Vec<&str>>());
 
                 let row_groups = parquet_sources_to_row_groups(scan_task.sources.as_slice());
-                let mp = read_parquet_into_micropartition(
+                read_parquet_into_micropartition(
                     uris.as_slice(),
                     columns.as_deref(),
                     None,
@@ -424,9 +424,7 @@ impl MicroPartition {
                     Some(schema.clone()),
                     field_id_mapping,
                 )
-                .context(DaftCoreComputeSnafu)?;
-
-                mp.cast_to_schema(schema).context(DaftCoreComputeSnafu)
+                .context(DaftCoreComputeSnafu)
             }
 
             // CASE: Last resort fallback option

src/daft-scan/src/lib.rs

@@ -253,7 +253,9 @@ impl DataFileSource {
 pub struct ScanTask {
     pub sources: Vec<DataFileSource>,
 
-    /// Schema to use when reading the DataFileSources.
+    /// Schema to use when reading the DataFileSources. This should always be passed in by the
+    /// ScanOperator implementation and should not have had any "pruning" applied.
+    ///
     /// Note that this is different than the schema of the data after pushdowns have been applied,
     /// which can be obtained with [`ScanTask::materialized_schema`] instead.
     pub schema: SchemaRef,