Fix integer overflow in combinations size-hint tests

[00200200]

Fixes #995.

combinations_inexact_size_hints panics on 32-bit because the test helper computes binomial coefficients with a naive product-then-divide. That overflows usize even when the result fits (C(27, 7) on 32-bit; C(21, 19) even on 64-bit).

Both integration-test helpers now call the existing checked_binomial implementation through a #[doc(hidden)] re-export, as requested in review. The overflow regression remains in place. The wrappers use unwrap() so an actually unrepresentable coefficient still fails the test rather than silently becoming zero.

Verification

On macOS aarch64 with Rust 1.98.1:

• cargo test --all-features --quiet: 446 integration tests and 210 documentation tests passed.
• cargo test --lib --all-features test_checked_binomial --quiet: passed.
• cargo check --no-default-features --quiet: passed.
• cargo check --no-default-features --features use_alloc --quiet: passed.
• RUSTDOCFLAGS=-Dwarnings cargo doc --all-features --no-deps --quiet: passed.
• cargo fmt --all -- --check and git diff --check: passed.

Strict Clippy (RUSTFLAGS="--deny warnings" cargo clippy --all-targets --all-features --quiet) reports four existing clippy::question_mark warnings outside the changed code. Reproduced the same four warnings on the previous PR head, 3de765066ea0fbfa770bb5d609f722588e981c43; this follow-up does not modify unrelated code or suppress those lints.

32-bit execution, Rust 1.63 MSRV, and Miri were not run locally.

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.
✅ Project coverage is 93.82%. Comparing base (6814180) to head (e6fce47).
⚠️ Report is 223 commits behind head on master.

Additional details and impacted files

@@            Coverage Diff             @@
##           master    #1131      +/-   ##
==========================================
- Coverage   94.38%   93.82%   -0.57%     
==========================================
  Files          48       52       +4     
  Lines        6665     6491     -174     
==========================================
- Hits         6291     6090     -201     
- Misses        374      401      +27     

☔ View full report in Codecov by Harness.
📢 Have feedback on the report? Share it here.

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.

[phimuemue]

If it's the same algorithm, can't we just call checked_binomial?

[00200200]

Because checked_binomial is pub(crate) inside the private adaptors module, it isn't visible to the integration test crates in tests/.

If you'd prefer to share the implementation across tests rather than keeping the local helper, I can expose it with #[doc(hidden)] pub use crate::adaptors::checked_binomial; in lib.rs and call it here as itertools::checked_binomial(n, k).unwrap_or(0). Would you like me to do that?

[phimuemue]

If it's really supposed to be the same code. I very much prefer this.

Or, alternatively, move the tests to src. I do not even know if we have a good reason to keep them under a separate directory, especially if we have [test] within src already.

@jswrenn opinions on this?

[00200200]

Addressed in e6fce47: both integration-test helpers now reuse checked_binomial through a #[doc(hidden)] re-export. All 656 integration/doc tests passed, along with the helper unit test, no_std checks, docs, and formatting. Strict Clippy reports the same four pre-existing question_mark warnings as the previous PR head.