The underrun behaviour of the USRP2 is different from the USRP1, and
the adaptive latency mechanism is not directly transferable. Instead,
fix the latency with a higher starting value, which effectively
buffers more samples on the host in front of the Ethernet interface.
An alternative may be to use the adaptive approach with USRP2
specific upper and lower bounds. For now, just use preprocessor
directives until a better solution comes around.
Signed-off-by: Thomas Tsou <ttsou@vt.edu>