libc/amd64: fix stpncpy.S again

[fuz]

Summary

The previous fix introduced a regression on machines without the BMI1
instruction set extension. The TZCNT instruction used in this function
behaves different on old machines when the source operand is zero, but
the code was originally designed to never trigger this case. The bug
fix caused this case to be possible, leading to a regression on
sufficiently old hardware.

Fix the code by messing with things such that the source operand is
never zero.

PR: 291720
Fixes: 66eb78377bf109af1d9e25626bf254b4369436ec

Test Plan

passes unit test. The behaviour on older machines was
manually checked by replacing TZCNT with BSF in the affected code
path. Please check again before I land this.

[markj]

Approved. If this doesn't land soon, say in the next day or two, we should revert the original change until it does.

[fuz]

In D54303#1241050, @markj wrote:

Approved. If this doesn't land soon, say in the next day or two, we should revert the original change until it does.

A change to completely disable the SIMD code until this is fixed (line 39) had already been committed when the regression was discovered, so the tree is currently in working order.

[cy]

I've been using this patch for a couple of weeks. No problems so far.