Add #[inline] to mul, sub, inversion and xgcd ops (#1246)

Follow-up to #1229, continuing the inlining work from #981.

This adds `#[inline]` to `mul`, `sub`, `inversion`, and `xgcd` where
benchmarks showed clear wins. I also tried inlining `mul_mod_special`,
but it ended up being significantly slower (~2×), so I left it as-is.

## Benchmarks

Criterion, 100 samples. Showing deltas outside noise:

| Benchmark | Change |
| --- | --- |
| `sub`, I512 | **-19.2%** |
| `xgcd/256` | **-3.9%** |
| `xgcd/3` | **-3.6%** |
| `concatenating_mul`, I128 | **-3.4%** |
| `sub`, I4096 | **-3.4%** |
| `xgcd/64` | **-3.1%** |
| `wrapping_square`, U256 | **-1.7%** |
| `wrapping_mul`, U256 | **-1.5%** |
| `wrapping_mul`, I4096 | **-1.5%** |
| `xgcd/16` | **-1.5%** |

No regressions outside noise.

Author: cong-or

src/int/mul.rs

@@ -49,6 +49,7 @@ impl<const LIMBS: usize> Int<LIMBS> {
     }
 
     /// Multiply `self` by `rhs`, returning a concatenated "wide" result.
+    #[inline]
     #[must_use]
     pub const fn concatenating_mul<const RHS_LIMBS: usize, const WIDE_LIMBS: usize>(
         &self,

src/uint/gcd.rs

@@ -221,6 +221,7 @@ impl<const LIMBS: usize> OddUint<LIMBS> {
     /// Execute the Extended GCD algorithm.
     ///
     /// Given `(self, rhs)`, computes `(g, x, y)` s.t. `self * x + rhs * y = g = gcd(self, rhs)`.
+    #[inline]
     #[must_use]
     pub const fn xgcd(&self, rhs: &Self) -> OddUintXgcdOutput<LIMBS> {
         OddUintXgcdOutput::from_pattern_output(self.binxgcd_odd(rhs))

src/uint/invert_mod.rs

@@ -105,6 +105,7 @@ impl<const LIMBS: usize> Uint<LIMBS> {
     ///
     /// If the inverse does not exist (`k > 0` and `self` is even, or `k > Self::BITS`),
     /// returns `CtOption::none`, otherwise returns `CtOption::some`.
+    #[inline]
     #[must_use]
     pub const fn invert_mod2k_vartime(&self, k: u32) -> CtOption<Self> {
         if k == 0 {
@@ -132,6 +133,7 @@ impl<const LIMBS: usize> Uint<LIMBS> {
     ///
     /// If the inverse does not exist (`k > 0` and `self` is even, or `k > Self::BITS`),
     /// returns `CtOption::none`, otherwise returns `CtOption::some`.
+    #[inline]
     #[must_use]
     pub const fn invert_mod2k(&self, k: u32) -> CtOption<Self> {
         let (odd, is_odd) = self.to_odd_or_one();

src/uint/mul.rs

@@ -10,6 +10,7 @@ pub(crate) mod schoolbook;
 
 impl<const LIMBS: usize> Uint<LIMBS> {
     /// Multiply `self` by `rhs`, returning a concatenated "wide" result.
+    #[inline]
     #[must_use]
     pub const fn concatenating_mul<const RHS_LIMBS: usize, const WIDE_LIMBS: usize>(
         &self,
@@ -121,6 +122,7 @@ impl<const LIMBS: usize> Uint<LIMBS> {
     }
 
     /// Square self, returning a concatenated "wide" result.
+    #[inline]
     #[must_use]
     pub const fn concatenating_square<const WIDE_LIMBS: usize>(&self) -> Uint<WIDE_LIMBS>
     where