feat(contracts): runtime-checked requires preconditions + counterexample evidence (v1.7.0, Theme A-lite)

[escapeboy]

Sprint 2 / Theme A-lite — agentlanguages.dev competitive-borrow program

Borrowed from Vera/Aver (design-by-contract) and Vow (counterexample = concrete replayable input). No SMT/Z3 — consistent with the 1.5.0 Decided ruling against symbolic model checking; stays in Boruna's concrete-trace + replay philosophy. (User chose this lighter path over full Z3.)

What

• requires <expr> preconditions compiled to runtime guards at function entry (reusing the previously-dormant Op::Assert — no bytecode bump). Functions without contracts emit no guard.
• Violation → VmError::ContractViolation { message, counterexample }, where counterexample is the offending arguments (positional, rendered) — the replayable input an auditor needs. Captured in the run's hash-chained audit log (tamper-evident).
• Stable error_kind contract_violation (retry: no — deterministic in inputs).
• ensures postconditions deferred (documented).

Gates (local, clippy 1.97)

compiler ✓ · vm ✓ · orchestrator ✓ (new contract + classify tests) · clippy --all-targets --features serve ✓ · fmt ✓.

Release

1.6.0 → 1.7.0 (CHANGELOG, README badge). See docs/design-contracts-runtime.md.

[github-actions]

Bench compare

Threshold for regression: ≥ 10% slower mean.

Benchmark	Mean change	99% CI
compile_crud_admin_template	-2.51%	[-5.64%, +0.58%]
compile_medium_program	+0.93%	[-0.92%, +2.41%]
compile_small_program	+2.49%	[+1.15%, +3.88%]
evidence_build_5_steps	+21.97%	[+12.54%, +31.85%]
evidence_build_empty	+16.64%	[+8.47%, +25.72%]
evidence_verify_10_steps	+113.68%	[+88.17%, +138.23%]
evidence_verify_5_steps	+70.37%	[+42.98%, +98.31%]
vm_call_dispatch_loop/iters=1000	-14.49%	[-18.83%, -9.67%]
vm_call_dispatch_loop/iters=10000	-14.17%	[-18.50%, -9.53%]
vm_pure_loop/iters=1000	-13.93%	[-17.08%, -10.54%]
vm_pure_loop/iters=10000	-14.64%	[-17.92%, -11.16%]
vm_pure_loop/iters=100000	+3.83%	[+1.61%, +5.91%]
vm_record_loop/iters=1000	+2.21%	[-0.65%, +4.90%]
vm_record_loop/iters=10000	+1.43%	[-2.17%, +5.03%]

⚠️ 4 benchmark(s) regressed past threshold:

• evidence_build_5_steps
• evidence_build_empty
• evidence_verify_10_steps
• evidence_verify_5_steps