TFHE-rs v1.3: Faster Division on CPU, Key Upgrader & Memory Tracking on GPU

TFHE-rs v1.3 introduces significant enhancements for Fully Homomorphic Encryption (FHE) workloads, with improvements spanning CPU, GPU, and HPU backends. Key updates include a fourfold boost in GPU performance for integer logarithm operations and up to 20% speed increases for other GPU tasks, alongside new utility functions for predicting GPU memory usage to prevent out-of-memory errors. CPU-side advancements feature a 36% faster division and an optimized Zero-Knowledge Proof (ZK Proof) verification process, which is now 34–38% quicker due to a new hashing scheme. The update also introduces noise squashing compression to manage large ciphertexts in Multi-Party Computation (MPC) settings, a key upgrader for updating old ciphertexts, and improved HPU backend capabilities with faster FPGA reloading and additional supported operations. These enhancements aim to bolster the performance, security, and usability of TFHE-rs, ensuring computational correctness with a failure probability below 2^-128.