Optimization: Improve conference mix loop performance

[ming]

When mixing hundreds or more ports, the mix loop is one of the bottlenecks. This ticket will reduce the CPU load substantially.

Thanks to Kal Conley (b17c0de) for the patch.

PJMEDIA test result using mips_test with 256 calls:

Clock  Item                                      Time     CPU    MIPS
 Rate                                           (usec)    (%)

without patch:
8KHz conference bridge with 256 calls            7391    0.739   59.88

with patch:
8KHz conference bridge with 256 calls            4888    0.489   39.60

[ming]

In 5687:

Fixed #2057: Optimization: Improve conference mix loop performance

[nanang]

In 5698:

Re #2057: Fixed compile error on MSVC2005 due to var declaration not in the beginning of block (C89 interop).