fix some trivial noise issues on some architecture with hidden markov models

[was]

From John Cremona:

Here's a test failure with sage-5.0.beta5 which I don't think I have seen before. Build from scratch on 32-bit ubuntu, gcc 4.4.3.

john@john-laptop%./sage -t  --long -force_lib
"devel/sage/sage/stats/hmm/chmm.pyx"
sage -t --long -force_lib "devel/sage/sage/stats/hmm/chmm.pyx"
**********************************************************************
File "/home/john/sage-5.0.beta5/devel/sage/sage/stats/hmm/chmm.pyx", line 865:
   sage: m.baum_welch([-2,-1,.1,0.1])
Expected:
   (4.534646052182663, 7)
Got:
   (4.534646052182662, 7)
**********************************************************************
File "/home/john/sage-5.0.beta5/devel/sage/sage/stats/hmm/chmm.pyx", line 867:
   sage: m.log_likelihood([-2,-1,.1,0.1])
Expected:
   4.534646052182663
Got:
   4.534646052182662
**********************************************************************
1 items had failures:
  2 of  21 in __main__.example_19
***Test Failed*** 2 failures.
For whitespace errors, see the file /home/john/.sage//tmp/chmm_19036.py
        [2.7 s]

[cremona]

Applied the the previously offending build of 5.0.beta5, this works fine.