Remove lossy casting in logspace

Currently `logspace` does a lossy cast from `F::Int` to `usize`. This
could be problematic in the rare cases that this is called with a step
count exceeding what is representable in `usize`.

Resolve this by instead adding bounds so the float's integer type itself
can be iterated.

library/compiler-builtins/libm/crates/libm-test/src/gen/domain_logspace.rs

@@ -1,5 +1,7 @@
 //! A generator that produces logarithmically spaced values within domain bounds.
 
+use std::ops::RangeInclusive;
+
 use libm::support::{IntTy, MinInt};
 
 use crate::domain::HasDomain;
@@ -34,6 +36,7 @@ pub fn get_test_cases<Op>(_ctx: &CheckCtx) -> impl Iterator<Item = (Op::FTy,)>
 where
     Op: MathOp + HasDomain<Op::FTy>,
     IntTy<Op::FTy>: TryFrom<usize>,
+    RangeInclusive<IntTy<Op::FTy>>: Iterator,
 {
     let domain = Op::DOMAIN;
     let start = domain.range_start();

library/compiler-builtins/libm/crates/libm-test/src/num.rs

@@ -1,8 +1,9 @@
 //! Helpful numeric operations.
 
 use std::cmp::min;
+use std::ops::RangeInclusive;
 
-use libm::support::{CastInto, Float};
+use libm::support::Float;
 
 use crate::{Int, MinInt};
 
@@ -214,7 +215,10 @@ fn as_ulp_steps<F: Float>(x: F) -> Option<F::SignedInt> {
 /// to logarithmic spacing of their values.
 ///
 /// Note that this tends to skip negative zero, so that needs to be checked explicitly.
-pub fn logspace<F: FloatExt>(start: F, end: F, steps: F::Int) -> impl Iterator<Item = F> {
+pub fn logspace<F: FloatExt>(start: F, end: F, steps: F::Int) -> impl Iterator<Item = F>
+where
+    RangeInclusive<F::Int>: Iterator,
+{
     assert!(!start.is_nan());
     assert!(!end.is_nan());
     assert!(end >= start);
@@ -225,7 +229,7 @@ pub fn logspace<F: FloatExt>(start: F, end: F, steps: F::Int) -> impl Iterator<I
     steps = steps.min(between); // At maximum, one step per ULP
 
     let mut x = start;
-    (0..=steps.cast()).map(move |_| {
+    (F::Int::ZERO..=steps).map(move |_| {
         let ret = x;
         x = x.n_up(spacing);
         ret