Fix size_of_hasher test for 32-bit Windows

Due to ABI differences between different 32-bit targets, the
`size_of_hasher` test wrongly failed on `i686-pc-windows-msvc`.

Although the test case with that name was introduced in GitoxideLabs#1915, the
failure is actually long-standing, in that an analogous faiure
occurred in the old `size_of_sha1` test that preceded it and on
which it is based. That failure only happened when the old
`fast-sha1` feature was enabled, and not with the old `rustsha1`
feature. It was not detected earlier as that target is irregularly
tested, and built with `--no-default-features --features max-pure`
more often than other targets due to difficulties building some
other non-Rust dependencies for it. Since GitoxideLabs#1915, the failure
happens more often, since we now use only one SHA-1 implementation,
`sha1-checked`, so the test always fails on `i686-pc-windows-msvc`.

This changes the test to use `gix_testtools::size_ok`, which makes
a `==` comparison on 64-bit targets but a `<=` comparison on 32-bit
targets where there tends to be more variation in data structures'
sizes. This is similar to the fixes in GitoxideLabs#1687 (77c3c59, fc13fc3).

Author: EliahKagan

gix-hash/tests/hash/hasher.rs

@@ -1,12 +1,15 @@
 use gix_hash::{Hasher, ObjectId};
+use gix_testtools::size_ok;
 
 #[test]
 fn size_of_hasher() {
-    assert_eq!(
-        std::mem::size_of::<Hasher>(),
-        if cfg!(target_arch = "x86") { 820 } else { 824 },
-        "The size of this type may be relevant when hashing millions of objects,\
-        and shouldn't change unnoticed. The DetectionState alone clocks in at 724 bytes."
+    let actual = std::mem::size_of::<Hasher>();
+    let expected = 824;
+    assert!(
+        size_ok(actual, expected),
+        "The size of this type may be relevant when hashing millions of objects, and shouldn't\
+        change unnoticed: {actual} <~ {expected}\
+        (The DetectionState alone clocked in at 724 bytes when last examined.)"
     );
 }