Misc coding styles and niche optimizations

[toyboot4e]

Newer style

• Replace return with pure
• Replace fmap with <$>

Associated modulus values

• GHC.TypeLits vs GHC.TypeNats
  Prefer the latter, which requires less type conversion from Nat.
• KnownNat vs AllowAmbiguousTypes vs Tagged
  Prefer KnownNat for simplicity.

Performance

• Index iteration: VU.forM_ (VU.generate n id) vs forM_ [0 .. n - 1]
  Latter was a bit faster on AtCoder, even beter than to Bundle, although there's slow for_ warnings.
• corece vs newtype constructors
  Prefer newtype constructors that helps us with typing. There's no performance difference (probably).
• Reverse vector iteration: for [n - 1, n - 2 .. 0] vs foldr'
• Folding sequence: vector (VU.foldM'_ f s (VU.generate n id)) vs list (foldM_ f s [0 .. n - 1])
• Vector initialization: replicate + copy, unsafeNew + copy + manual filling vs grow -> manual filling