[PATCH] Fix glFenceSync() retracing

[Ryan C. Gordon]

> I see the problem, but unfortunately this change will break retracing
> glFlushMappedBufferRange with a sub-range, and user memory pointers as
> recorded by the virtual-address-regions branch (which tries to
> reconstruct data in user arrays via mallocs and memcpys).

I was worried there would be side effects like this. In the short term, 
it's not hard for me to use GL_NV_fence when tracing, so this isn't a 
big emergency for me; a better long-term solution in apitrace is the 
right decision, it was beyond my limited understanding of the codebase, 
though.

--ryan.