Fix panic on --runs 0

[VXNCXNX]

What's broken

--runs 0 and --max-runs 0 panic with an arithmetic overflow instead of reporting the bad value. --min-runs 0 panics too, for any command slower than the minimum benchmarking time.

Repro

$ hyperfine --runs 0 'true'
Benchmark 1: true
thread 'main' panicked at src/benchmark/mod.rs:277:31:
attempt to subtract with overflow

Exit code is 101. It is easy to hit from a script where the run count comes from a variable.

The fix

Benchmark::run computes count - 1 after the initial timing run, so a count of zero underflows. An explicit maximum of zero is rejected in Options::from_cli_arguments, since asking for at most zero runs cannot be satisfied. A zero minimum is treated as valid and clamped to 1 instead of rejected: no lower bound is a coherent request, and the initial run has already happened by then. Happy to reject --min-runs 0 as well if you would rather have it symmetric.

Verification

$ hyperfine --runs 0 'true'
Error: At least one run has to be performed. Please specify a value larger than zero for '--runs'
$ hyperfine --debug-mode --min-runs 0 'sleep 4'
Benchmark 1: sleep 4
  Time (abs ≡):         4.000 s   [User: 0.000 s, System: 0.000 s]

Two tests added in tests/integration_tests.rs. Full suite passes (44 / 19 / 41), cargo fmt --check clean.

---

Disclosure: this patch was written with the help of Claude (an LLM). I reviewed the diff, ran the test suite, and verified the behaviour against a real build before opening this.