*   Change embedding op to use parallel version of `DynamicStitch`.
    *   Add support for sparse multidimensional feature columns.
    *   Speed up the case for sparse float columns that have only 1 value.
    *   Allow sparse float splits to support multivalent feature columns.
    *   Add `quantile` to `tf.distributions.TransformedDistribution`.
    *   Add `NCHW_VECT_C` support for `tf.depth_to_space` on GPU.

Software // Implementations of Modular Exponentiation" by Shay Gueron // [https://eprint.iacr.org/2011/239.pdf]. var c uint for i := 0; i < n; i++ { _ = T[n+i] // bounds check elimination hint // Step 1 (T = a × b) is computed as a large pen-and-paper column // multiplication of two numbers with n base-2^_W digits. If we just // wanted to produce 2n-wide T, we would do // // for i := 0; i < n; i++ { // d := bLimbs[i] // T[n+i] = addMulVVW(T[i:n+i], aLimbs, d) // } // // where d is a digit of the multiplier,...

The sequence of execution before FastAPI 0.106.0 was like this diagram:

Time flows from top to bottom. And each column is one of the parts interacting or executing code.

```mermaid
sequenceDiagram

participant client as Client
participant handler as Exception handler
participant dep as Dep with yield

Software // Implementations of Modular Exponentiation" by Shay Gueron // [https://eprint.iacr.org/2011/239.pdf]. var c uint for i := 0; i < n; i++ { _ = T[n+i] // bounds check elimination hint // Step 1 (T = a × b) is computed as a large pen-and-paper column // multiplication of two numbers with n base-2^_W digits. If we just // wanted to produce 2n-wide T, we would do // // for i := 0; i < n; i++ { // d := bLimbs[i] // T[n+i] = addMulVVW(T[i:n+i], aLimbs, d) // } // // where d is a digit of the multiplier,...