data_stats() table size undercounts blob v2 data

Description

Blob v2 stores data in a different place so dataset.stats.data_stats() only counts the size of the pointer to it, not the full data. But when we calculate a table's size, usually (I think) we want to include the size of all the data involved.

Steps to reproduce

import pathlib, shutil
import pyarrow as pa
import lance
from lance.blob import BlobType, blob_field

URI = "/tmp/blob_size_mre"
PAYLOAD = 1 << 20  # 1 MiB -> past dedicated_size_threshold, so a sidecar

shutil.rmtree(URI, ignore_errors=True)

blob_type = BlobType()
field = blob_field("blob", dedicated_size_threshold=1024)
storage = pa.array(
    [{"data": b"x" * PAYLOAD, "uri": None, "position": None, "size": None}],
    type=blob_type.storage_type,
)
table = pa.table(
    [pa.ExtensionArray.from_storage(blob_type, storage)], schema=pa.schema([field])
)
ds = lance.write_dataset(table, URI, data_storage_version="2.2")

reported = sum(f.bytes_on_disk for f in ds.stats.data_stats().fields)
on_disk = sum(p.stat().st_size for p in pathlib.Path(URI).rglob("*") if p.is_file())
print(f"payload written   : {PAYLOAD:>9,}")
print(f"data_stats total  : {reported:>9,}")
print(f"actual bytes used : {on_disk:>9,}")

Expected behavior

data_stats should return about 1Mb

Lance version

9.0.0

Language binding

Python

Environment

local osx

Logs / traceback

Start with the supplied Python reproduction and the data_stats() implementation, focusing on how Blob v2 sidecar files are accounted for. Compare reported bytes_on_disk with the total file sizes, then verify that data_stats() includes the full blob payload and reports about 1 MiB.