feat: add 256-bit A/B arithmetic, hashing-fee hook, passthrough cleanup

[QuickMythril]

Summary

The planned "jar-fork batch" — the three AT-VM features that require changes inside the jar (unlike the map functions, which ride the platform passthrough range). One commit, 4f70a18.

1. A/B 256-bit arithmetic (0x0140–0x0147)

The eight spec-reserved ops (Add_A_To_B … Div_B_By_A) that CIYAM's at_api.html documents and Signum implemented (cases 320–327 in AtApiController.java); Qortal never did. A1..A4/B1..B4 are treated as one 256-bit little-endian value, matching Signum's limb order. Add/sub/mul wrap modulo 2^256 and are bit-identical to Signum's results.

Two deliberate divergences from Signum, both documented in javadoc:

• Division is unsigned (Signum's is signed — differs only when a top bit is set; a test pins our behavior).
• Divide-by-zero throws IllegalOperationException → machine error / onErrorAddress, matching this library's existing DIV_DAT/DIV_VAL/MOD_DAT convention (Signum silently no-ops).

2. Hashing-fee discrimination

New API.getHashingFunctionSteps(OpCode, FunctionCode, MachineState) hook + FunctionCode.isHashingFunction() (codes 0x0200–0x0207), consulted by MachineState.execute() for the hash built-ins. The CIYAM spec explicitly recommends charging more for hash functions; no chain could because the pricing path never saw the FunctionCode. Default delegates down the existing chain, so with no platform override, charged steps are byte-for-byte identical — including for consumers that override only the general 3-arg getOpCodeSteps (pinned by test).

3. Platform passthrough cleanup

The ad-hoc 0x0500–0x06ff literals in FunctionCode.valueOf become documented PLATFORM_CODE_START/PLATFORM_CODE_END constants + isPlatformFunctionCode(int), with the dispatch contract (precheck + execute) spelled out in API_PASSTHROUGH javadoc. valueOf behavior is unchanged for every input (boundary-tested).

Backward compatibility (consensus-critical)

• 0x0140–0x0147 were previously unmapped → fatal IllegalFunctionCodeException, so no deployed bytecode can contain them on a live execution path; existing codes untouched.
• Hashing pricing defaults are byte-identical without an explicit platform override.
• Passthrough change is refactor + docs only.
• Nothing reaches qortium-core until it deliberately bumps its pinned ciyam-at.version (currently 33df17d). The Core consumer (ChainFunctionCode/ChainATAPI) was checked and needs no changes.

Verification

• mvn -B clean verify: 207 tests, 0 failures, 0 errors (baseline on main: 179) — verified independently in a clean run, not just the implementation pass.
• 28 new tests: 18 arithmetic (per-op direction/result register, carry/borrow across all four limbs, wrap, cross-limb mul/div, truncating + unsigned-vs-signed division discriminator, div-by-zero both directions), 6 hashing-hook (default unchanged, override scope, delegation-chain compatibility), 4 passthrough (range predicates, valueOf boundaries, dispatch of an unenumerated code, wrong-signature precheck rejection).
• 3 mutation checks (ADD→SUB, hook bypass, silent div-by-zero) each made the new tests fail before restoration.

Merge note

Squash is fine here (nothing pins this commit). Core repin is intentionally deferred — this batch should not ride the pre-70,000 release; repin when a feature actually consumes it.

🤖 Generated with Claude Code