docs(readme): refresh top-N numbers at 80M rows for consistency

Top-N (first 10/100 rows) re-measured at the same 80M-row OHLC file
used by the full-scan table. File size affects open + footer scan
overhead, so numbers differ from the 10M run:
- top-10:  Java 2,701 vs JNI 550 ops/s → 4.9x
- top-100: Java 2,697 vs JNI 573 ops/s → 4.7x

Both tables now reference the same fixture.

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

Author: dfa1

README.md

@@ -26,8 +26,8 @@ ops/s = complete file scans per second; higher is better.
 
 | N rows | Java (ops/s) | JNI/Rust (ops/s) | Speedup |
 |--------|-------------|-----------------|---------|
-| 10     | 5,739       | 1,618           | **3.5×** |
-| 100    | 5,787       | 1,660           | **3.5×** |
+| 10     | 2,701       | 550             | **4.9×** |
+| 100    | 2,697       | 573             | **4.7×** |
 
 Measured 2026-06-16, commit `74ec207b`. See [docs/explanation.md](docs/explanation.md#benchmarks) for full tables and methodology.