This lets us retry each failed test.
This is beneficial because the gnupg race is sort-of independent.
When we want to test every commit, the success probability before
is
( 1 - P(gnupg causes test fail)) ) ^ (number of tests)
for each commit. With this change it's
( 1 - P(gnupg causes test fail)^retries ) ^ (number of tests)
which is much much better.
Signed-off-by: Ian Jackson <ijackson@chiark.greenend.org.uk>