runtime: reentrant nanotime and walltime may return bogus time on ARM

[prattmic]

https://golang.org/cl/246763 made nanotime and walltime reentrant, but on ARM, the calculation of the result used a register expected to be zero which was changed by https://golang.org/cl/246763 to the old value of vdsoSP. This is usually 0, but in a reentrant call would not be, thus generating a bogus result.

I already fixed walltime in https://golang.org/cl/361159, but nanotime has a similar issue.

cc @cherrymui

[gopherbot]

Change https://golang.org/cl/362674 mentions this issue: runtime: add upper half and carry bit with zero