Adds read_complex_slice (#23): a bulk decode fast path that reads a BEVE
complex array (the bytes produced by write_complex_slice / to_vec_complex_slice)
into a Vec<Complex<T>> in a single bounds-checked copy, instead of the
element-by-element serde visitor path. ~21x faster on a 4096-element
Complex<f64> decode.

Purely additive (one new public function, no removals, no signature changes,
no new trait surface), so this is a minor bump.