binary tracing

[Martin Hunt]

I am implementing step 1, just replacing the _stp_printf with one that
can write binary suing "%b". Numbers look good so far.

S : printf 5 integers            3926    2134    1633
S : printf 10 integers           5667    3054    2325
S : binary printf 5 integers     1937    1170     880
S : binary printf 10 integers    2031    1203     925

I hacked up another test which says using the translator optimization
Tom proposed could cut up to 150ns off the 5 integer results. Transport
optimizations should cut those a bit further. Keep in mind, these
include kprobe overhead which is

S : empty probe                  1575     789     589

Martin