test(parquet): measure imported Vortex file size vs source Parquet

Adds taxiParquet_importedSize_vsOriginal — imports NYC Yellow Taxi
2024-01 (~3M rows, 19 columns, mix of I64 / F64 / I32 / Utf8) via
ParquetImporter and prints the size ratio. Uses the same cached parquet
file as ParquetVsVortexReadBenchmark so no extra download in normal runs.

Current measurement (after b4d1b43 added global dict for Utf8):

  Parquet  = 49,961,641 bytes (47.6 MB)
  Vortex   = 49,260,172 bytes (47.0 MB)
  ratio    = 0.99x

Vortex is now 1% smaller than Parquet on this dataset. The improvement
over the prior 50 MB baseline tracks the new Utf8 global-dict path —
taxi has several low-cardinality string-style columns whose per-chunk
dicts were duplicated wastefully before.

Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>

Author: dfa1

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

@@ -155,6 +155,38 @@ void stringColumnValuesMatch(@TempDir Path tmp) throws Exception {
         }
     }
 
+    @Test
+    void taxiParquet_importedSize_vsOriginal(@TempDir Path tmp) throws Exception {
+        // Given — NYC Yellow Taxi 2024-01 (~3M rows, 19 cols, mix of I64 / F64 / I32 / Utf8).
+        // Uses the same parquet file as ParquetVsVortexReadBenchmark (cached in /tmp).
+        Path cached = Path.of("/tmp", "yellow_tripdata_2024-01.parquet");
+        Path src;
+        if (java.nio.file.Files.exists(cached)) {
+            src = cached;
+        } else {
+            assumeNetworkAvailable();
+            src = tmp.resolve("yellow_tripdata_2024-01.parquet");
+            try (var in = URI.create("https://d37ci6vzurychx.cloudfront.net/trip-data/yellow_tripdata_2024-01.parquet")
+                    .toURL().openStream()) {
+                java.nio.file.Files.copy(in, src);
+            }
+        }
+        Path vortex = tmp.resolve("taxi.vortex");
+
+        // When
+        ParquetImporter.importParquet(src, vortex);
+
+        // Then
+        long parquetSize = java.nio.file.Files.size(src);
+        long vortexSize = java.nio.file.Files.size(vortex);
+        System.out.printf(
+                "[TaxiSizeComparison] Parquet=%,d bytes (%.1f MB)  Vortex=%,d bytes (%.1f MB)  Vortex/Parquet=%.2fx%n",
+                parquetSize, parquetSize / 1_048_576.0,
+                vortexSize, vortexSize / 1_048_576.0,
+                (double) vortexSize / parquetSize);
+        assertThat(vortexSize).isGreaterThan(0);
+    }
+
     @Test
     void parquetAndVortexRowCountsAreEqual(@TempDir Path tmp) throws Exception {
         // Given