Before:
$ time ./fend 2^1555.555|b3sum
0567ac5de698ba87f79145ebf29c1ab169d37b185e76ddd4b8ecc2de2c82d4de  -

real	1m42,334s
user	1m41,908s
sys	0m0,032s

Afer:
$ time ./fend 2^1555.555|b3sum
0567ac5de698ba87f79145ebf29c1ab169d37b185e76ddd4b8ecc2de2c82d4de  -

real	0m0,401s
user	0m0,401s
sys	0m0,004s

And for the gcd:

Before:
$ time ./fend 0.9911^192|b3sum
315e6e5e5b9716413c0e6cbe347ed8a7e2923b3cec6290db7d88ff9710aea09e  -

real	0m19,810s
user	0m19,757s
sys	0m0,005s

After:
$ time ./fend 0.9911^192|b3sum
315e6e5e5b9716413c0e6cbe347ed8a7e2923b3cec6290db7d88ff9710aea09e  -

real	0m0,049s
user	0m0,040s

System information: rust 1.92.0, LLVM 20.1.8 x86_64-redhat-linux-gnu,
Intel i5-13600K, opt-level=3, ~/.cargo/config.toml contents:
[build]
rustflags = ["-C", "target-cpu=native"]

Functions changed:

add_assign_internal: Replaces high-level get/set overhead with
efficient, vectorizable slice iteration using .zip() to eliminate bounds
checks in the hot loop.

sub: Optimizes borrowing logic using u128 overflow detection and safely
clamps loop ranges to handle unnormalized inputs without panicking.

sub_assign: Performs subtraction in-place on mutable buffers to
eliminate expensive memory allocations during the Karatsuba
recombination step.

mul: Dispatches to Karatsuba multiplication for inputs larger than 64
limbs (4096 bits) to reduce algorithmic complexity from
O(N²) to O(N^¹·⁵⁸⁵).

mul_karatsuba_slice: Recursively computes Karatsuba products using
&[u64] slices instead of cloning BigUint vectors, significantly reducing
memory churn.

mul_internal_slice: Provides a highly optimized Schoolbook (O(N²))
multiplication base case that operates directly on slices for maximum
cache efficiency.

div_rem_knuth (called by divmod): Replaces slow bit-by-bit binary
division with Knuth's Algorithm D (base-2⁶⁴), reducing the number of
division steps by a factor of 64.

root_n: Replaces Binary Search (linear convergence) with Newton's Method
(quadratic convergence), reducing iteration count from thousands to
dozens for large roots.

lshift / rshift: Moves interrupt checks outside the loop and simplifies
carry logic to allow the compiler to generate efficient block memory
moves.

add_assign_shifted: A specialized helper for Karatsuba that performs a
"shift-and-add" operation in one pass without creating intermediate
shifted values.

gcd: Replaces the division-heavy Euclidean algorithm with Stein's
Algorithm (Binary GCD), utilizing efficient bitwise shifts and
subtraction to eliminate expensive modulo operations during fraction
simplification.