Add mod dependencies for minor optimisations of prime fields

[Multramate]

Pull Request Checklist

• [x] Update ChangeLog.md
• [x] Tests are passing.
• [x] Benchmarks are not regressing.

[sdiehl]

I think this is fine. What is the delta in performance with and without unsafeCoerce?

[Multramate]

I think this is fine. What is the delta in performance with and without unsafeCoerce?

Without unsafeCoerce, we would have to use fromInteger, which does unnecessary slow modular reductions. Using the current benchmarks on my machine,

• addition takes ~45ns before this PR, ~35ns with unsafeCoerce, ~105ns with fromInteger, and
• multiplication takes ~145ns before this PR, ~135ns with unsafeCoerce, ~205ns with fromInteger.

I did not measure the other operations but their deltas are relatively negligible (e.g. division already takes ~1us, so nanosecond boosts are unnoticeable). Essentially, there is a minor boost in performance in all fronts, but with higher robustness due to abstracting low-level optimisations completely to mod.

Alternatively, we could request for Mod to be exposed in Data.Mod.Unsafe, for instance.

[sdiehl]

Ok sounds good to me.