Remove spin dependency on x86 and x86_64 targets.

[briansmith]

Here's a representative example of how this changes the generated assembly in callers:

@@ -36,14 +36,11 @@
 	lea r12, [rcx - 4]
3343
 		Acquire => intrinsics::atomic_load_acquire(dst),
-	mov rax, qword ptr [rip + ring::cpu::intel::featureflags::FEATURES@GOTPCREL]
-	movzx eax, byte ptr [rax + 4]
-		match self.status.load(Ordering::Acquire) {
-	cmp al, 2
-		if let Some(value) = self.get() {
-	jne .LBB85_2
+	mov rax, qword ptr [rip + ring::cpu::intel::featureflags::FEATURES]
+		match NonZeroUsize::new(val) {
+	test rax, rax
+	je .LBB85_2
 	movabs rax, 274877906881
 		if input.len() > MAX_IN_OUT_LEN {
@@ -104,10 +101,9 @@
 		if in_out.len() >= SSE_MIN_LEN {
 	cmp r12, 129
 	jb .LBB85_8
-		*features
-	mov rax, qword ptr [rip + ring::cpu::intel::featureflags::FEATURES@GOTPCREL]
-	mov eax, dword ptr [rax]
3343
+		Acquire => intrinsics::atomic_load_acquire(dst),
+	mov rax, qword ptr [rip + ring::cpu::intel::featureflags::FEATURES]
 		if (self.values() & MASK) == MASK {
 	test eax, 256
@@ -216,9 +212,9 @@
 .LBB85_2:
 	.cfi_def_cfa rbp, 16
 	mov r13, r8
-		self.try_call_once_slow(f)
-	call spin::once::Once<T,R>::try_call_once_slow
+		None => self.init(f),
+	call ring::polyfill::once_cell::race::OnceNonZeroUsize::init
 	mov r8, r13
 	movabs rax, 274877906881

[codecov]

Codecov Report

Attention: Patch coverage is 90.38462% with 5 lines in your changes missing coverage. Please review.

Project coverage is 96.63%. Comparing base (be72a40) to head (96e2676).
Report is 5 commits behind head on main.

Files with missing lines	Patch %	Lines
src/polyfill/once_cell/race.rs	82.14%	5 Missing ⚠️

Additional details and impacted files

@@           Coverage Diff           @@
##             main    #2331   +/-   ##
=======================================
  Coverage   96.63%   96.63%           
=======================================
  Files         175      176    +1     
  Lines       21551    21597   +46     
  Branches      526      527    +1     
=======================================
+ Hits        20825    20871   +46     
- Misses        611      613    +2     
+ Partials      115      113    -2     

☔ View full report in Codecov by Sentry.
📢 Have feedback on the report? Share it here.

[briansmith]

This was rebased on top of #2338 and #2337 to eliminate some tiny regressions in CPU feature dispatching performance due to reloads.

The new version incorporates matklad/once_cell#273. PR #2338 seems to obviate the need for adding get_unchecked but I did start the process of working w/ upstream on that at matklad/once_cell#274. I also filed a documentation clarification at rust-lang/rust#136669 to support the once_cell change.