Hack to support forcing normalization (#7517)

## Summary

We need to support normalization of vector arrays to perform faster
compute. However, this will essentially always create larger arrays on
disk. This is a hack specifically for `L2Denorm` that will allow us
write normalized vector arrays to disk.

Given that writing scalar fn arrays to disk is already gated by an
environment variable and the unstable encodings feature flag, I think
this hack is fine for now.

## Testing

This just allows some arrays that are larger than canonical.

Signed-off-by: Connor Tsui <connor.tsui20@gmail.com>

Author: connortsui20

vortex-compressor/src/compressor.rs

@@ -28,6 +28,7 @@ use vortex_array::arrays::list::ListArrayExt;
 use vortex_array::arrays::listview::ListViewArrayExt;
 use vortex_array::arrays::listview::list_from_list_view;
 use vortex_array::arrays::primitive::PrimitiveArrayExt;
+use vortex_array::arrays::scalar_fn::AnyScalarFn;
 use vortex_array::arrays::struct_::StructArrayExt;
 use vortex_array::dtype::DType;
 use vortex_array::dtype::Nullability;
@@ -254,6 +255,11 @@ impl CascadingCompressor {
                     return Ok(result);
                 }
 
+                // TODO(connor): HACK TO SUPPORT L2 DENORMALIZATION!!!
+                if result.is::<AnyScalarFn>() {
+                    return Ok(result);
+                }
+
                 // Otherwise, fall back to compressing the underlying storage array.
                 let compressed_storage = self.compress(ext_array.storage_array())?;
 
@@ -337,6 +343,11 @@ impl CascadingCompressor {
                 // TODO(connor): Add a tracing warning here too.
                 return Ok(compressed);
             }
+
+            // TODO(connor): HACK TO SUPPORT L2 DENORMALIZATION!!!
+            if compressed.is::<AnyScalarFn>() {
+                return Ok(compressed);
+            }
         }
 
         // No scheme improved on the original.