Quinn test failure regression from fastbloom 0.15

``` ---- bloom_token_log::test::k_num_zero stdout ---- thread 'bloom_token_log::test::k_num_zero' (5526) panicked at /home/runner/.cargo/registry/src/index.crates.io-1949cf8c6b5b557f/fastbloom-0.16.0/src/builder.rs:129:1: attempt to subtract with overflow note: run with `RUST_BACKTRACE=1` environment variable to display a backtrace ``` This test fails in a way that seems to be caused by upgrading fastbloom from 0.14 to 0.15 (or 0.16). Is this intentional? ```rust fn test_doesnt_panic(log: BloomTokenLog) { let mut rng = new_rng(); let issued = SystemTime::now(); let lifetime = Duration::from_secs(1_000_000); for _ in 0..200 { let _ = log.check_and_insert(rng.random::<u128>(), issued, lifetime); } } #[test] fn k_num_zero() { test_doesnt_panic(BloomTokenLog::new(100, 0)); } ``` See: - https://github.com/quinn-rs/quinn/pull/2540