`tf.data.Options()`.
    *   Support cancellation of long-running `reduce`.
    *   Now we use `dataset` node name as prefix instead of the op name, to
        identify the component correctly in metrics, for pipelines with repeated
        components.
    *   Improve the performance of datasets using `from_tensors()`.
    *   Promoting `unbatch` from experimental to core API.

// | | // | ... | // | | // +--------- // The common components(signatureR, signatureS, hashed message, public key X, // and public key Y) each takes block size of bytes. The block size is different for // different curves and is set by canUseKDSA function. var params [4096]byte // Copy content into the parameter block. In the verify case, // we copy signature (r), signature(s), hashed message, public key x component, // and public key y component into the parameter block. p := params[:0] p = appendBlock(p,...

This release adds support for [Trio](https://trio.readthedocs.io/en/stable/). ✨

It upgrades the version of Starlette to `0.15.0`, now based on [AnyIO](https://anyio.readthedocs.io/en/stable/), and the internal async components in **FastAPI** are now based on AnyIO as well, making it compatible with both **asyncio** and **Trio**.