Split the dsound test

[Robert Reif]

Francois Gouget wrote:

>
> You will notice the following problems:
>  * the test takes awfully long (>6 minutes here). It loops through the 
> known formats and for each of them it plays, in turn, the regular 
> test, a test through a duplicated buffer, with a listener, etc.
>  * as a result of the above, a user listening to the test does not 
> know what to expect, which makes it hard to know whether the sound was 
> played correctly. 


The corresponding Microsoft test takes hours to complete.  If time
is an issue, then the test should be split up into a quick test and
an exhaustive test.

Also, volume and pan tests on primary and secondary buffers is
not the same thing as 3D.  They just happen to be similar now
because we don't support hardware acceleration yet.