fix: remove azure suffix range request (delta-io#1006)

resolves delta-io#968 (I hope)

## What changes are proposed in this pull request?
TLDR: Fixes kernel (default-engine) error when trying to read on azure.

`arrow-55` introduced [support for reading parquet metadata via suffix
range requests](apache/arrow-rs#7334) but since
azure [doesn't support suffix range
requests](https://github.com/apache/arrow-rs-object-store/blob/9c43521d3c8e86fdfc4df788a5a63e45e23093b7/src/azure/client.rs#L912-L918)
whenever anyone tried reading an azure table with default-engine based
on arrow-55 it [blew
up](delta-io#968).

This PR mitigates by re-adding the original `head` request so that we
can pass the `file_size` to the `ParquetObjectReader::with_file_size`
which states that when this API is used, the reader will ensure only
bounded range requests are used:
```rust
/// Provide the byte size of this file.
///
/// If provided, the file size will ensure that only bounded range requests are used. If file
/// size is not provided, the reader will use suffix range requests to fetch the metadata.
///
/// Providing this size up front is an important optimization to avoid extra calls when the
/// underlying store does not support suffix range requests.
///
/// The file size can be obtained using [`ObjectStore::list`] or [`ObjectStore::head`].
pub fn with_file_size(self, file_size: usize) -> Self { ... }
```

## How was this change tested?
tested manually.. painfully manually. spun up a public azure container
thing and pointed our example code at it. was able to repro the
following error without my changes and observe it going away after
changes here applied.
```
Parquet(
    External(
        NotSupported {
            source: "Azure does not support suffix range requests",
        },
    ),
)
```

obviously highlights our test gap. working on setting up some
environments to test against S3/ADLS/GCS/etc.

Author: zachschuermann

kernel/src/engine/default/parquet.rs

@@ -265,7 +265,28 @@ impl FileOpener for ParquetOpener {
 
         Ok(Box::pin(async move {
             #[cfg(feature = "arrow-55")]
-            let mut reader = ParquetObjectReader::new(store, path);
+            let mut reader = {
+                use crate::object_store::ObjectStoreScheme;
+                // HACK: unfortunately, `ParquetObjectReader` under the hood does a suffix range
+                // request which isn't supported by Azure. For now we just detect if the URL is
+                // pointing to azure and if so, do a HEAD request so we can pass in file size to the
+                // reader which will cause the reader to avoid a suffix range request.
+                // see also: https://github.com/delta-io/delta-kernel-rs/issues/968
+                //
+                // TODO(#1010): Note that we don't need this at all and can actually just _always_
+                // do the `with_file_size` but need to (1) update our unit tests which often
+                // hardcode size=0 and (2) update CDF execute which also hardcodes size=0.
+                if let Ok((ObjectStoreScheme::MicrosoftAzure, _)) =
+                    ObjectStoreScheme::parse(&file_meta.location)
+                {
+                    // also note doing HEAD then actual GET isn't atomic, and leaves us vulnerable
+                    // to file changing between the two calls.
+                    let meta = store.head(&path).await?;
+                    ParquetObjectReader::new(store, path).with_file_size(meta.size)
+                } else {
+                    ParquetObjectReader::new(store, path)
+                }
+            };
             #[cfg(all(feature = "arrow-54", not(feature = "arrow-55")))]
             let mut reader = {
                 // TODO avoid IO by converting passed file meta to ObjectMeta (no longer an issue