Generic Backtracking

Generic implementation of backtracking together with some examples. This came out of a code dojo session with some of my colleagues (https://github.com/pacman82/cobra-kai-code-dojo). I found it neat enough to put it into its own repository.

---

Fork additions

This fork adds test coverage (the library and all five examples now have unit tests; CI runs cargo test --all-targets against both the default and parallel feature sets) and two backward-compatible extensions to the Problem/Solutions API:

• Problem::Possibility now requires only Clone, not Copy. Every Copy type is already Clone, so existing implementations are unaffected; the weaker bound additionally allows heap-backed decision types (an owned String, a Vec, ...) without forcing them through an artificial handle/index indirection just to satisfy the trait.
• An opt-in, experimental parallel feature adds parallel_solutions. It searches the independent branches rooted at each top-level possibility in parallel (via rayon), with each branch still walked sequentially by an ordinary Solutions iterator on its own thread. The extra Clone + Send + Sync bounds only apply at that new call site — Solutions::new and its bounds are untouched, and the rayon dependency is not compiled unless the feature is enabled. Newer and less exercised than the core API; the word_chain example (cargo run --example word_chain --release --features parallel) uses it to time a word-chain search sequentially and in parallel side by side. Its shape may still change in a minor version.