Concurrent create dataset gives bad error, doesn't retry

[wjones127]

If a user tries to append to the same dataset concurrently, one writer will fail with a confusing internal error:

Internal { message: "Commit conflict", location: Location { file: "rust/lance-table/src/io/commit.rs", line: 424, column: 27 } }

This is because in commit_new_dataset, we don't retry to catch the CommitError::CommitConflict:

lance/rust/lance/src/io/commit.rs

Lines 126 to 138 in d8da445

	write_manifest_file(
	object_store,
	commit_handler,
	base_path,
	&mut manifest,
	if indices.is_empty() {
	None
	} else {
	Some(indices.clone())
	},
	write_config,
	)
	.await?;

For comparison, in commit_transaction(), we have a retry loop and a nicer error message:

lance/rust/lance/src/io/commit.rs

Lines 486 to 494 in d8da445

	Err(crate::Error::CommitConflict {
	version: target_version,
	source: format!(
	"Failed to commit the transaction after {} retries.",
	commit_config.num_retries
	)
	.into(),
	location: location!(),
	})

We should change this to see if the transactions are compatible and, if so, have commit_new_dataset load the dataset and call into commit_transaction.