fix(contracts): fuzz the real burn entrypoint, not a reimplementation

[hman38705]

Summary

fuzz_burn.rs previously reimplemented burn's arithmetic inline (saturating_sub on plain i128s), never calling the actual TokenFactory::burn entrypoint — described accurately elsewhere in this same fuzz suite as a "pure reimplementation." No target imported soroban_sdk or TokenFactory at all, since the contract crate wasn't even a fuzz dependency (fuzz/Cargo.toml only listed libfuzzer-sys and arbitrary).

• Adds contracts/token-factory/testutils-lib, a thin rlib-only crate that compiles the contract's own src/lib.rs so it can be depended on as an ordinary library. This is a separate crate rather than adding "rlib" to the contract's own crate-type — I tried that first and it broke the production WASM build's fat-LTO size optimization (measured locally: ~42 KB → ~68 KB, over the 64 KB budget scripts/check-wasm-size.mjs enforces). The real contract package's Cargo.toml is unchanged (crate-type = ["cdylib"]).
• Adds TokenFactory::fuzz_seed_token, a #[cfg(feature = "testutils")]-only plain associated function (not a #[contractimpl] entrypoint — compiled out of the production build, invisible on-chain, documented in docs/contract-abi.md's new "Test/fuzz-only helpers" section) that registers a token in factory storage exactly as create_token would have. It uses a real Stellar Asset Contract as the token — the same bypass the crate's own seed_token test helper already uses, since no real token WASM exists anywhere in this repo to install at a token_wasm_hash in a native test/fuzz environment (see the existing note on mod bench in lib.rs).
• fuzz_burn.rs now drives the real TokenFactory::burn entrypoint via the generated TokenFactoryClient against a live soroban_sdk::testutils::Env, asserting each result against an independently-computed oracle rather than reimplementing burn's own logic.

Scope: this covers fuzz_burn.rs, the simplest of the flagged targets. fuzz_fee_arithmetic.rs (plus auditing fuzz_mint_tokens.rs/fuzz_set_metadata.rs for the same pattern, and new targets for create_tokens_batch/distribute_fee/migrate) are left for follow-up — each needs live fee-split registration and multi-recipient balance tracking, a substantially larger harness than burn's single-token case.

closes #1098

Test plan

• cargo fmt -- --check — clean
• cargo test (contracts/token-factory) — 184/184 passed
• cargo clippy -- -D warnings — clean
• cargo build --target wasm32v1-none --release — WASM size unchanged at 41,670 bytes (limit 65,536)
• All 5 fuzz targets (cargo +nightly build --release) still build, including the 4 untouched ones
• Ran fuzz_burn for 159,607 iterations against the real, correct contract — zero crashes/false positives
• Deliberately weakened burn's amount > balance check locally, confirmed the rewritten target catches it (the underlying token rejects the resulting over-balance burn and the fuzz target panics on the unexpected result) within the first few thousand runs, then reverted the regression before committing
• scripts/check-abi-doc-drift.sh, scripts/check-validation-drift.sh, node scripts/check-vercel-cron.mjs all pass
• Frontend test suite (npm run test -- --run) — 702/702 passed

[github-actions]

🎉 This PR is included in version 1.5.9 🎉

The release is available on GitHub release

Your semantic-release bot 📦🚀