Two changes since 1.5.0, both landed on main:

- read_typed_slice (#24): a bulk decode fast path that reads a BEVE typed
  numeric array (the bytes produced by write_typed_slice / to_vec_typed_slice /
  to_writer_typed_slice) into a Vec<T> in a single bounds-checked copy, instead
  of the element-by-element serde visitor path. ~4.75x faster on a 4096-element
  f64 decode. The read counterpart of the existing typed-slice writers; also
  ships dev/unify-bulk-slice-unsafe.md, a deferred plan to unify the bulk-slice
  unsafe primitive across readers and writers.

- Complex<f16> / Complex<bf16> serde support (#25): a correctness fix. The bulk
  complex path already handled half floats, but serde did not -- missing
  Serialize impls, a deserializer that rejected float byte_code 0/1, and a
  payload-size formula that mis-sized bf16. Half-float complex values now
  round-trip through to_vec / from_slice.

Additive plus a bug fix, no removals or signature changes, so this is a minor
bump.