Fix Parquet read with an isNull filter on nested array (#10890)

Summary:
Below failure occurs when selecting a struct column from Parquet filtering
using an isNull filter on nested array and the filter is extracted to subfield
filters in 'extractFiltersFromRemainingFilter'.

```
velox/dwio/parquet/reader/PageReader.cpp:737, Function:skip, Expression:  No decoder to skip, Source: RUNTIME, ErrorCode: INVALID_STATE
unknown file: Failure
C++ exception with description "Exception: VeloxRuntimeError
Error Source: RUNTIME
Error Code: INVALID_STATE
Reason: No decoder to skip
```

Pull Request resolved: #10890

Reviewed By: Yuhta

Differential Revision: D62885150

Pulled By: kevinwilfong

fbshipit-source-id: a940c092f03a6dbf234312e05306760cf3556c26

velox/dwio/parquet/reader/PageReader.cpp

@@ -719,8 +719,14 @@ void PageReader::skip(int64_t numRows) {
   }
   firstUnvisited_ += numRows;
 
+  if (toSkip == 0) {
+    return;
+  }
   // Skip nulls
   toSkip = skipNulls(toSkip);
+  if (toSkip == 0) {
+    return;
+  }
 
   // Skip the decoder
   if (isDictionary()) {

velox/dwio/parquet/tests/reader/ParquetTableScanTest.cpp

@@ -370,7 +370,6 @@ TEST_F(ParquetTableScanTest, decimalSubfieldFilter) {
       "Scalar function signature is not supported: eq(DECIMAL(5, 2), DECIMAL(5, 1))");
 }
 
-// Core dump is fixed.
 TEST_F(ParquetTableScanTest, map) {
   auto vector = makeMapVector<StringView, StringView>({{{"name", "gluten"}}});
 
@@ -399,7 +398,6 @@ TEST_F(ParquetTableScanTest, nullMap) {
   assertSelectWithFilter({"i", "c"}, {}, "", "SELECT i, c FROM tmp");
 }
 
-// Core dump is fixed.
 TEST_F(ParquetTableScanTest, singleRowStruct) {
   auto vector = makeArrayVector<int32_t>({{}});
   loadData(
@@ -414,7 +412,6 @@ TEST_F(ParquetTableScanTest, singleRowStruct) {
   assertSelectWithFilter({"s"}, {}, "", "SELECT (0, 1)");
 }
 
-// Core dump and incorrect result are fixed.
 TEST_F(ParquetTableScanTest, array) {
   auto vector = makeArrayVector<int32_t>({});
   loadData(
@@ -528,6 +525,21 @@ TEST_F(ParquetTableScanTest, reqArrayLegacy) {
       "SELECT UNNEST(array[array['a', 'b'], array[], array['c', 'd']])");
 }
 
+TEST_F(ParquetTableScanTest, filterOnNestedArray) {
+  loadData(
+      getExampleFilePath("struct_of_array.parquet"),
+      ROW({"struct"},
+          {ROW({"a0", "a1"}, {ARRAY(VARCHAR()), ARRAY(INTEGER())})}),
+      makeRowVector(
+          {"unused"},
+          {
+              makeFlatVector<int32_t>({}),
+          }));
+
+  assertSelectWithFilter(
+      {"struct"}, {}, "struct.a0 is null", "SELECT ROW(NULL, NULL)");
+}
+
 TEST_F(ParquetTableScanTest, readAsLowerCase) {
   auto plan = PlanBuilder(pool_.get())
                   .tableScan(ROW({"a"}, {BIGINT()}), {}, "")