fix: flush non-blocking log appender on shutdown instead of mem::forget

[orrinfrazier]

Closes #12

Summary

• init_logging leaked the tracing_appender WorkerGuard via mem::forget, so buffered log lines were never guaranteed to flush on shutdown — losing the final lines that matter for crash/shutdown post-mortems.
• Return the WorkerGuard from init_logging (now #[must_use]) and hold it in main() for the program lifetime, so it drops — and flushes the file appender — after the final shutdown log line.

Changes

• logging.rs: extract the appender construction into a private create_file_appender(config) -> (NonBlocking, WorkerGuard); change init_logging to return WorkerGuard; remove mem::forget(guard) and its import.
• main.rs: bind let _log_flush_guard = init_logging(&config); before the tokio runtime so it drops last on scope exit (after drop(rt) and info!("Shutdown complete.")).
• lib.rs: update the doc example to hold the guard.
• Add a unit test (file_appender_flushes_on_guard_drop) that writes a line, drops the guard, and asserts the line was flushed to a temp log file.

Testing

• cargo test -p cuprated --lib — 12 passed, 0 failed (includes the new flush test)
• cargo clippy -p cuprated --all-targets -- -D warnings — clean
• cargo fmt -p cuprated -- --check — clean
• cargo build --bin cuprated — builds

Notes

• No new dependencies (tempfile was already a dev-dependency, tracing-appender already a dependency).
• The guard is intentionally held in main rather than Node: init_logging runs before Node::launch, and the terminal Shutdown complete. line is emitted in main after the node/runtime drop, so the guard must outlive Node.

Refs: issues/S10-1-log-flush-shutdown.md