Use serde_derive via derive feature on serde

[waywardmonkeys]

No description provided.

[dtolnay]

See #155 — the way serde is currently used is intentional.

Thanks anyway for the PR!

[waywardmonkeys]

Interesting! I'm not seeing nearly that on my machine (an M3 Pro Max).

time reports about 10 to 10.2 seconds (total time) on a cargo build in this crate, but the difference comes in with the parallelization (as expected) as the utilization is 328% with them separate and 228% with this PR (actual wall time differs by about 1.5 seconds).

[dtolnay]

Wall time is the relevant value.

On my machine (2990WX) it's 5.7 seconds before this PR and 9.1 seconds after this PR, which is a 60% regression.