Fix a couple of possible divide-by-zero conditions in statistics

Several people reported the average time problem, but there's another
one lurking there too: if the dive duration is zero, you get bogus
average depth information too (but because that one was a floating point
divide, and by default they are unsignalling on x86, it didn't crash, it
just resulted in bogus results).

Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>