docs: update benchmark table with latest results on M1 Pro

Zolt now beats Jolt across all 14 programs — large SHA-256 workloads
improved from ~1.0x parity to 0.53-0.71x.

Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>

Author: MatteoMer

README.md

@@ -77,20 +77,20 @@ Compared against [Jolt](https://github.com/a16z/jolt) (Rust) on Apple M1 Pro, 16
 
 | Program | Trace | Jolt (ms) | Zolt (ms) | Ratio |
 |---------|------:|----------:|----------:|------:|
-| fibonacci | 128 | 702.21 | 195.80 | 0.28x |
-| factorial | 64 | 877.07 | 166.29 | 0.19x |
-| bitwise | 512 | 763.48 | 235.37 | 0.31x |
-| collatz | 2048 | 888.86 | 382.80 | 0.43x |
-| primes | 2048 | 860.59 | 421.58 | 0.49x |
-| sum | 16 | 669.59 | 157.24 | 0.23x |
-| gcd | 256 | 690.98 | 255.23 | 0.37x |
-| signed | 16 | 682.64 | 160.60 | 0.24x |
-| primes_large | 65536 | 1729.97 | 1757.10 | 1.02x |
-| sha256_128 | 16384 | 1476.15 | 1468.76 | 0.99x |
-| sha256 | 8192 | 1256.52 | 1215.65 | 0.97x |
-| sha256_512 | 32768 | 2056.07 | 2084.61 | 1.01x |
-| sha256_1024 | 65536 | 2947.85 | 2886.95 | 0.98x |
-| sha256_2048 | 131072 | 3917.25 | 4029.60 | 1.03x |
+| fibonacci | 128 | 700.95 | 157.25 | 0.22x |
+| factorial | 64 | 686.91 | 139.28 | 0.20x |
+| bitwise | 512 | 769.48 | 205.33 | 0.27x |
+| collatz | 2048 | 867.76 | 281.20 | 0.32x |
+| primes | 2048 | 840.17 | 303.15 | 0.36x |
+| sum | 16 | 652.35 | 131.52 | 0.20x |
+| gcd | 256 | 682.48 | 193.43 | 0.28x |
+| signed | 16 | 672.60 | 143.37 | 0.21x |
+| primes_large | 65536 | 1769.51 | 1260.96 | 0.71x |
+| sha256_128 | 16384 | 1464.11 | 903.51 | 0.62x |
+| sha256 | 16384 | 1822.50 | 964.19 | 0.53x |
+| sha256_512 | 32768 | 1992.00 | 1268.11 | 0.64x |
+| sha256_1024 | 65536 | 2712.83 | 1848.69 | 0.68x |
+| sha256_2048 | 131072 | 4669.83 | 3191.04 | 0.68x |
 
 ## Upstream Compatibility