[Data] Avoid pickling LanceFragment when creating read tasks for Lance

[c21]

Why are these changes needed?

Avoid pickling LanceFragment when creating read tasks for Lance, as this is expensive.

Related issue number

Checks

• I've signed off every commit(by using the -s flag, i.e., git commit -s) in this PR.
• I've run scripts/format.sh to lint the changes in this PR.
• I've included any doc changes needed for https://docs.ray.io/en/master/.
  • I've added any new APIs to the API Reference. For example, if I added a
    method in Tune, I've added it in doc/source/tune/api/ under the
    corresponding .rst file.
• I've made sure the tests are passing. Note that there might be a few flaky tests, see the recent failures at https://flakey-tests.ray.io/
• Testing Strategy
  • Unit tests
  • Release tests
  • This PR is not tested :(

[wjones127]

I think you can still keep these. We've established that count_rows is not the slow part. In fact, it's even faster than get_fragments().

Also, small nit: you don't need the [] inside of sum. If you omit them you get a generator expression which bypasses the need to allocate the whole list.

num_rows = sum(f.count_rows() for f in fragments)

[c21]

Yes, updated.

[wjones127]

If you wanted something that did some IO prefetching, you could instead do:

Suggested change

	for fragment_id in fragment_ids:
	fragment = lance_ds.get_fragment(fragment_id)
	batches = fragment.to_batches(columns=columns, filter=row_filter)
	for batch in batches:
	yield pyarrow.Table.from_batches([batch])
	fragments = [lance_ds.get_fragment(id) for id in fragment_ids]
	scanner = lance_ds.scanner(
	columns,
	filter=row_filter,
	fragments=fragments,
	)
	for batch in scanner.to_reader():
	yield pyarrow.Table.from_batches([batch])

[c21]

Cool, updated.