Optimize EIP-196 AltBn128 EcAdd and EcMul

[siladu]

PR description

Pull in changes from besu-eth/besu-native#301

Key Improvements:

• 🚀 EcAdd (average cases): Up to 2.56x faster (149% throughput increase)
• 🚀 EcAdd (worst cases): 1.3-1.4x faster (32-36% throughput increase)
• ✅ EcMul (scalar multiplication average cases) : ~8-9% faster (thanks to gnark-crypto v0.19.2)
• ✅ Pairings: Slightly improved (~0.6%)

  | Operation    | Before (ns) | After (ns) | Speedup  | Throughput Gain |
  |--------------|-------------|------------|----------|-----------------|
  | EcAdd        | 1,618.3     | 649.8      | 2.49x 🚀 | +138%           |
  | EcAddMarius  | 4,173.7     | 3,133.5    | 1.33x    | +33%            |
  | EcAddAmez1   | 4,019.2     | 3,053.7    | 1.32x    | +32%            |
  | EcAddAmez2   | 4,055.4     | 3,046.6    | 1.33x    | +33%            |
  | EcAddAmez3   | 4,074.4     | 3,081.2    | 1.32x    | +31%            |
  | EcAddCase0   | 4,185.6     | 3,024.6    | 1.38x    | +34%            |
  | EcAddCase1   | 4,040.7     | 2,942.1    | 1.37x    | +36%            |
  | EcAddCase100 | 1,584.8     | 618.4      | 2.56x 🚀 | +148%           |
  | EcAddCase106 | 1,412.5     | 561.4      | 2.52x 🚀 | +149%           |
  | mul1         | 51,909.7    | 48,001.8   | 1.08x    | +8%             |
  | mul2         | 51,479.8    | 47,235.8   | 1.09x    | +9%             |
  | 2 pairings   | 442,882.6   | 440,668.6  | 1.01x    | +0.6%           |
  | 4 pairings   | 628,526.7   | 624,919.0  | 1.01x    | +0.6%           |
  | 6 pairings   | 814,240.0   | 809,483.1  | 1.01x    | +0.6%           |

benchmark executed on m6a.2xlarge - 4c/8t 32 GB

More detailed results: besu-eth/besu-native#301

Testing

• passes besu reference tests
• passes enhanced eest ecadd tests feat(tests): add test "framework" for BN254 precompiles ethereum/execution-specs#1871
• hoodi full sync on AMD
• hoodi partial full sync on ARM (smoke test)
• mainnet test (SNAP then left to run)

Profiling

Before PR on the left, this PR on the right - shows main improvement is removal of IntByReference for tracking output size and parsing error strings which were only logged at TRACE level. In this PR, we log an error code at DEBUG level, which can be looked up if necessary. Avoiding passing Strings across the native boundary is also an improvement.

[Copilot]

Copilot encountered an error and was unable to review this pull request. You can try again by re-requesting a review.

[ahamlat]

LGTM