sync: avoid a dynamic check in WaitGroup on 64-bit architectures

sync: avoid a dynamic check in WaitGroup on 64-bit architectures

uint64 is guaranteed by the compiler to be aligned on 64-bit archs.
By using uint64+uint32 instead of [3]uint32 we can make use of the
guaranteed alignment to avoid the run-time alignment check.

On linux/amd64:

name                     old time/op    new time/op    delta
WaitGroupUncontended-4     8.84ns ± 3%    7.62ns ± 4%  -13.72%  (p=0.000 n=17+18)
WaitGroupAddDone-4         66.8ns ± 3%    45.9ns ± 2%  -31.31%  (p=0.000 n=20+18)
WaitGroupAddDoneWork-4     79.2ns ± 1%    56.6ns ± 1%  -28.54%  (p=0.000 n=17+20)
WaitGroupWait-4            2.83ns ± 2%    2.58ns ± 2%   -9.05%  (p=0.000 n=18+20)
WaitGroupWaitWork-4        16.8ns ± 6%    16.5ns ± 6%     ~     (p=0.072 n=20+18)
WaitGroupActuallyWait-4     263ns ± 2%     261ns ± 5%     ~     (p=0.063 n=18+20)

Change-Id: I314340f2ed8a47d8b9c15f8a3b07e41f252f4831
Reviewed-on: https://go-review.googlesource.com/c/go/+/189837
Reviewed-by: Cherry Mui <cherryyz@google.com>
Reviewed-by: Ian Lance Taylor <iant@golang.org>
Run-TryBot: Cherry Mui <cherryyz@google.com>
TryBot-Result: Go Bot <gobot@golang.org>