Same millisecond generation is not monotonic

[richardhundt]

Version 3.1.0

>>> from ulid import ULID
>>> 
>>> prev = None
>>> for i in range(2_000_000):
...     u = str(ULID())
...     if prev is not None and u < prev:
...         print(f"drop at {i}:")
...         print(f"  prev={prev}  ts={prev[:10]} rand={prev[10:]}")
...         print(f"  curr={u}  ts={u[:10]} rand={u[10:]}")
...         break
...     prev = u
... 
drop at 17:
  prev=01KT6JE36AYQD46B7SP5T0K6K9  ts=01KT6JE36A rand=YQD46B7SP5T0K6K9
  curr=01KT6JE36A9KNN12R9RC3JDJK0  ts=01KT6JE36A rand=9KNN12R9RC3JDJK0

You can see that the rand part's lexicographic sorting has gone down (9 < Y) and it bailed after 17 iterations.