Fix winmm's check_position() tests

[Robert Reif]

I would just revert the patch and recalculate duration based on the fragment
size.  You need to keep the double because of integer overflow in the MS
calculation and integer underflow in the SMPTE calculation.  We need to
add a ceil to the frame calculation to round up partial frames.

Francois Gouget wrote:

> Hi Robert,
>
> Robert Reif wrote:
>
>> Please revert this patch.  It is wrong and produces the following 
>> errors in windows xp:
>
>
> Aaaargh, you were right. The problem was caused by another change I 
> made a long time ago while experimenting with wave_generate_la(). I 
> had completely forgotten that this change added an extra sample for a 
> given duration so I looked everywhere but there. That also explains 
> why my results were consistent across platforms.