feat: support update_columns for fragment operator

[SaintBacchus]

Now the support fragment.merge_columns to add a new column

import lance
import pyarrow as pa
import pyarrow.compute as pc
table = pa.table({"a": [1, 2, 3, 4], "b": ["a", "b", "c", "d"]})
dataset = lance.write_dataset(table, "example")

dataset.to_table().to_pandas()

def double_a(batch: pa.RecordBatch) -> pa.RecordBatch:
    doubled = pc.multiply(batch["a"], 2)
    return pa.record_batch([doubled], ["a_doubled"])

fragments = []
for fragment in dataset.get_fragments():
    new_fragment, new_schema = fragment.merge_columns(double_a, columns=['a'])
    fragments.append(new_fragment)

operation = lance.LanceOperation.Merge(fragments, new_schema)
dataset = lance.LanceDataset.commit("example", operation,
                                    read_version=dataset.version)
dataset.to_table().to_pandas()

However many users want to update the values of the existing column. If we're going to archive this scenario, we should do three steps:

1. add a column using the merge columns API
2. delete the old column
3. rename the new column into the old one.

Maybe it's better to provide a new API named update_columns to do these in one step.

The update operator in dataset API may not be efficient for the large lance dataset.

[aheev]

I think #4715 implements this

[SaintBacchus]

Nice to implement it~