behavior.
    *   Adding `clear_losses` API to be able to clear losses at the end of
        forward pass in a custom training loop in eager.
    *   Add support for passing list of lists to the `metrics` param in Keras
        `compile`.
    *   Added top-k to precision and recall to keras metrics.
    *   Adding public APIs for `cumsum` and `cumprod` keras backend functions.

"domain separation" bits and the first bit of // the padding. Sections 6.1 and 6.2 of [1] separate the outputs of the // SHA-3 and SHAKE functions by appending bitstrings to the message. // Using a little-endian bit-ordering convention, these are "01" for SHA-3 // and "1111" for SHAKE, or 00000010b and 00001111b, respectively. Then the // padding rule from section 5.1 is applied to pad the message to a multiple // of the rate, which involves adding a "1" bit, zero or more "0" bits, and // a final "1"...