numericStringLess("1a", "11a") would strip the equal prefix "1" and
then compare the bytes "a" and "1", when it should have compared the
numbers 1 and 11. Fix it by handling the case where the last equal
byte was numeric and the first differing byte is numeric in one
string and non-numeric in the other.
numericStringLess("12", "101") would strip the equal prefix "1" and
then compare the numbers 2 and 01, when it should have compared the
numbers 12 and 101. Fix it by tracking the beginning of the sequence
of numeric bytes containing the differing byte.
Test: sort_test.go
Change-Id: I8d9252a64625ba6a3c75d09bb1429dcb1115e3e1
