Implements predicate pushdown in the scan path by filtering stripes
based on column statistics before reading data. This is the core
integration that enables I/O reduction.

Changes:
1. Modified OrcScanTask to accept selected stripe indices
   - Changed Execute() to read only selected stripes using ReadStripe()
   - Iterator returns one batch per stripe (stripe = unit of parallelism)

2. Modified OrcScanTaskIterator to call FilterStripes
   - Applies predicate pushdown automatically during scan
   - FilterStripes ensures metadata is loaded
   - Returns task with selected stripes (empty if none match)

3. Stripe-selective reading
   - Task reads selected stripes one at a time using ReadStripe()
   - Replaces previous GetRecordBatchReader() which read all stripes
   - Implements stripe-level granularity from Task #0.5

Benefits:
- Skips stripes where predicate is known to be unsatisfiable
- Skips empty stripes (num_rows == 0)
- Reduces I/O by avoiding reads of filtered stripes
- Statistics loaded lazily and cached incrementally

Example: Query "WHERE x > 1000" on file with 10 stripes
- Stripe 0: x in [0, 100] -> Skip (literal(false))
- Stripe 5: x in [500, 600] -> Skip (literal(false))
- Stripe 9: x in [900, 1100] -> Scan (may have x > 1000)
Result: Only scans stripes that may contain matching rows

Verified: Mirrors cpp/src/arrow/dataset/file_parquet.cc lines 619-636

Co-authored-by: Claude Sonnet 4.5 <[email protected]>

Implements predicate pushdown in the scan path by filtering stripes
based on column statistics before reading data. This is the core
integration that enables I/O reduction.

Changes:
1. Modified OrcScanTask to accept selected stripe indices
   - Changed Execute() to read only selected stripes using ReadStripe()
   - Iterator returns one batch per stripe (stripe = unit of parallelism)

2. Modified OrcScanTaskIterator to call FilterStripes
   - Applies predicate pushdown automatically during scan
   - FilterStripes ensures metadata is loaded
   - Returns task with selected stripes (empty if none match)

3. Stripe-selective reading
   - Task reads selected stripes one at a time using ReadStripe()
   - Replaces previous GetRecordBatchReader() which read all stripes
   - Implements stripe-level granularity from Task #0.5

Benefits:
- Skips stripes where predicate is known to be unsatisfiable
- Skips empty stripes (num_rows == 0)
- Reduces I/O by avoiding reads of filtered stripes
- Statistics loaded lazily and cached incrementally

Example: Query "WHERE x > 1000" on file with 10 stripes
- Stripe 0: x in [0, 100] -> Skip (literal(false))
- Stripe 5: x in [500, 600] -> Skip (literal(false))
- Stripe 9: x in [900, 1100] -> Scan (may have x > 1000)
Result: Only scans stripes that may contain matching rows

Verified: Mirrors cpp/src/arrow/dataset/file_parquet.cc lines 619-636

Co-authored-by: Claude Sonnet 4.5 <[email protected]>