runtime maps: random iteration and maps.clone

[dgryski]

This PR fixes a few map-related issues:

• make map iteration less predictable. It's still not a fully random key, but then again neither is Big Go's.
• fix building the tsip runtime hash. At some point this broken when an import cycle was introduced, so just inline the bits we need from math/bits and encoding/binary.
• add maps.clone so the maps package can use it.
• initialize fastrand() on process startup so we actually get random numbers for map seeds
• fix build for tsip memhash

[aykevl]

I find it ironic that big Go said map iteration order is undefined, so made map iteration random, and now there's a lot of code depending on the random iteration order.
(There's a difference between random and undefined! Always producing the map keys in sorted order would be entirely valid according to the Go spec).

Anyway, see my comment below. Didn't investigate the code that changes the iteration order in detail, but it seems reasonable on the surface.

[aykevl]

Good idea, I didn't think of reusing the growing code for map cloning.

[dgryski]

Need to fix the binary sizes again ..