test(integration): use cascading(3) in WriteFileSizeIntegrationTest for fair JNI comparison

Java writer was using WriteOptions.defaults() (no cascading), while JNI uses
ALP+FastLanes bitpacking by default — an apples-to-oranges comparison.
Switching to cascading(3) on the Java side brings the ratio from ~6x down to
~1.16x (Java=1.2 MB vs JNI=1.0 MB for 100K OHLC rows). Tighten assertion
to < 2x to catch regressions.

Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>

Author: dfa1

integration/src/test/java/io/github/dfa1/vortex/integration/WriteFileSizeIntegrationTest.java

@@ -82,7 +82,7 @@ class WriteFileSizeIntegrationTest {
 
 	private static void writeJava(Path file, List<OhlcGenerator.OhlcBatch> batches) throws IOException {
 		try (FileChannel ch = FileChannel.open(file, StandardOpenOption.CREATE, StandardOpenOption.WRITE);
-		     VortexWriter writer = VortexWriter.create(ch, JAVA_SCHEMA, WriteOptions.defaults())) {
+		     VortexWriter writer = VortexWriter.create(ch, JAVA_SCHEMA, WriteOptions.cascading(3))) {
 			for (OhlcGenerator.OhlcBatch b : batches) {
 				writer.writeChunk(Map.of(
 						"date", b.dates(), "open", b.open(), "high", b.high(),
@@ -200,8 +200,8 @@ void fileSizeComparison(@TempDir Path tmp) throws IOException {
 
 		assertThat(javaSize).isPositive();
 		assertThat(jniSize).isPositive();
-		// Java writer uses ALP (no further compression) for F64, PrimitiveEncoding for integers.
-		// JNI uses ALP+FastLanes delta+bitpacking. Bound < 6× catches gross regressions.
-		assertThat(javaSize).isLessThan(jniSize * 6);
+		// Both writers use cascading(3): ALP+bitpacking for F64, delta+bitpacking for I64.
+		// Bound < 2× catches regressions while tolerating minor encoding strategy differences.
+		assertThat(javaSize).isLessThan(jniSize * 2);
 	}
 }