ping: optimize random number generation

[marten-seemann]

1. We shouldn't use time-seeded random numbers. They contain very, very little entropy.
2. We only need to initialize a math/rand random number generator once per Ping invocation.
3. We don't need to use go-ipfs-util for this, bringing down the number of dependencies (therefore: part of After repo consolidation, prune deps #1646).

[MarcoPolo]

look good. Did you do any benchmark to measure the actual perf optimization here? I would expect time seeded randomness to be pretty fast so curious to see if there was any actual perf benefit.

[marten-seemann]

Initializing from crypto/rand is minimally slower, but then we only do it once per Ping.

BenchmarkTimeSeededRand-10        100039             11698 ns/op
BenchmarkCryptoRand-10             95367             11659 ns/op