Task #13: Implement FilterStripes function (apache#33)

Main entry point for stripe filtering with predicate pushdown.
Mirrors Parquet's FilterRowGroups pattern.

Implementation:
1. Ensure complete metadata is loaded (file, manifest, statistics cache)
2. Call TestStripes to evaluate predicate against stripe statistics
3. Filter results to include only stripes where:
   - Predicate is satisfiable (not literal(false))
   - Stripe is non-empty (num_rows > 0)
4. Return vector of selected stripe indices

Stripes are skipped if:
- The predicate simplifies to literal(false) given statistics
- The stripe contains zero rows

This function is called by:
- ScanBatchesAsync (for scan optimization)
- Subset (for fragment splitting)
- TryCountRows (for count optimization)

Verified: Mirrors cpp/src/arrow/dataset/file_parquet.cc FilterRowGroups (lines 918-931)

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

Author: cbb330

cpp/src/arrow/dataset/file_orc.cc

@@ -757,16 +757,62 @@ Status OrcFileFragment::ClearCachedMetadata() {
   return Status::OK();
 }
 
+/// \brief Filter stripes based on predicate and column statistics
+///
+/// Main entry point for stripe filtering. Ensures metadata is loaded,
+/// calls TestStripes to evaluate statistics, and returns indices of
+/// stripes that may contain matching rows.
+///
+/// Stripes are skipped if:
+/// - The predicate simplifies to literal(false) given the stripe's statistics
+/// - The stripe is empty (num_rows == 0)
+///
+/// This mirrors Parquet's FilterRowGroups pattern.
+///
+/// \param[in] predicate The filter expression to evaluate
+/// \return Vector of stripe indices that should be scanned
 Result<std::vector<int>> OrcFileFragment::FilterStripes(
     compute::Expression predicate) {
-  // TODO: Implement in Task #13
-  // For now, return all stripes
+  // Ensure all required metadata is loaded
   ARROW_RETURN_NOT_OK(EnsureCompleteMetadata());
 
+  // Use TestStripes to evaluate the predicate against stripe statistics
+  ARROW_ASSIGN_OR_RAISE(auto expressions, TestStripes(std::move(predicate)));
+
+  // Build list of selected stripe indices
   std::vector<int> selected_stripes;
+
+  // Determine which stripes we're evaluating
+  // If stripes_ is set, use those indices; otherwise use all stripes
   int64_t num_stripes = metadata_->NumberOfStripes();
-  for (int64_t i = 0; i < num_stripes; ++i) {
-    selected_stripes.push_back(static_cast<int>(i));
+  std::vector<int> stripes_to_check;
+  if (stripes_) {
+    stripes_to_check = *stripes_;
+  } else {
+    stripes_to_check.resize(num_stripes);
+    for (int64_t i = 0; i < num_stripes; ++i) {
+      stripes_to_check[i] = static_cast<int>(i);
+    }
+  }
+
+  // Filter based on statistics and empty stripes
+  DCHECK(expressions.empty() || (expressions.size() == stripes_to_check.size()));
+  for (size_t i = 0; i < expressions.size(); ++i) {
+    int stripe_idx = stripes_to_check[i];
+
+    // Skip if predicate is unsatisfiable for this stripe
+    if (!expressions[i].IsSatisfiable()) {
+      continue;
+    }
+
+    // Skip empty stripes
+    auto stripe_info = metadata_->GetStripeInformation(stripe_idx);
+    if (stripe_info.num_rows == 0) {
+      continue;
+    }
+
+    // This stripe should be scanned
+    selected_stripes.push_back(stripe_idx);
   }
 
   return selected_stripes;