context: reduce contention in cancelCtx.Done

Use an atomic.Value to hold the done channel.
Conveniently, we have a mutex handy to coordinate writes to it.

name                 old time/op  new time/op  delta
ContextCancelDone-8  67.5ns ±10%   2.2ns ±11%  -96.74%  (p=0.000 n=30+28)

Fixes #42564

Change-Id: I5d72e0e87fb221d4e230209e5fb4698bea4053c6
Reviewed-on: https://go-review.googlesource.com/c/go/+/288193
Trust: Josh Bleecher Snyder <josharian@gmail.com>
Trust: Sameer Ajmani <sameer@golang.org>
Run-TryBot: Josh Bleecher Snyder <josharian@gmail.com>
TryBot-Result: Go Bot <gobot@golang.org>
Reviewed-by: Bryan C. Mills <bcmills@google.com>

src/context/benchmark_test.go

@@ -5,6 +5,7 @@
 package context_test
 
 import (
+	"context"
 	. "context"
 	"fmt"
 	"runtime"
@@ -138,3 +139,17 @@ func BenchmarkCheckCanceled(b *testing.B) {
 		}
 	})
 }
+
+func BenchmarkContextCancelDone(b *testing.B) {
+	ctx, cancel := context.WithCancel(context.Background())
+	defer cancel()
+
+	b.RunParallel(func(pb *testing.PB) {
+		for pb.Next() {
+			select {
+			case <-ctx.Done():
+			default:
+			}
+		}
+	})
+}