ParquetMetaDataPushDecoder API to clear all buffered ranges (#9673)

This PR is a follow up for [this ticket
](#8676). Implement same API
but for the metadata decoder.

See also
#9624 (comment)

## Rationale for this change
`ParquetMetaDataPushDecoder` clears exact requested ranges, but larger
speculative pushed ranges can remain buffered in `PushBuffers`. This
adds a way for callers to explicitly release non exact ranges

## What changes are included in this PR?
This adds `clear_all_ranges()`, which clears all byte ranges still
staged in the decoder's internal `PushBuffers`
## Are these changes tested?
yes
## Are there any user-facing changes?
Yes, this adds a new public `clear_all_ranges()` API on
`ParquetMetaDataPushDecoder`

Author: nathanb9

parquet/src/file/metadata/push_decoder.rs

@@ -358,6 +358,11 @@ impl ParquetMetaDataPushDecoder {
         Ok(())
     }
 
+    /// Clear any staged byte ranges currently buffered for future decode work.
+    pub fn clear_all_ranges(&mut self) {
+        self.buffers.clear_all_ranges();
+    }
+
     /// Try to decode the metadata from the pushed data, returning the
     /// decoded metadata or an error if not enough data is available.
     pub fn try_decode(&mut self) -> Result<DecodeResult<ParquetMetaData>> {
@@ -573,6 +578,23 @@ mod tests {
         assert!(metadata.offset_index().is_some());
     }
 
+    #[test]
+    fn test_metadata_decoder_clear_all_ranges() {
+        let file_len = test_file_len();
+        let mut metadata_decoder = ParquetMetaDataPushDecoder::try_new(file_len).unwrap();
+
+        metadata_decoder
+            .push_range(test_file_range(), TEST_FILE_DATA.clone())
+            .unwrap();
+        assert_eq!(metadata_decoder.buffers.buffered_bytes(), test_file_len());
+
+        metadata_decoder.clear_all_ranges();
+        assert_eq!(metadata_decoder.buffers.buffered_bytes(), 0);
+
+        let ranges = expect_needs_data(metadata_decoder.try_decode());
+        assert_eq!(ranges, vec![test_file_len() - 8..test_file_len()]);
+    }
+
     /// Decode the metadata incrementally, simulating a scenario where exactly the data needed
     /// is read in each step
     #[test]

parquet/src/util/push_buffers.rs

@@ -156,7 +156,6 @@ impl PushBuffers {
     }
 
     /// Clear all buffered ranges and their corresponding data
-    #[cfg(feature = "arrow")]
     pub fn clear_all_ranges(&mut self) {
         self.ranges.clear();
         self.buffers.clear();