Follow Random API more closely

[devmotion]

#1314 defines Random.Sampler for Sampleables with sampler.

I think one could consider

• switching to Random.Sampler completely and deprecating sampler,
• committing to the Random interface more closely. E.g., Random uses rand!(rng, x, sampler) whereas Distributions supports rand!(rng, sampler, x). If desired, one could implement rand!(rng, x, sampler) as well and possibly deprecate rand!(rng, sampler, x).

[mschauer]

All agreed except that I think that is only an intermediate solution. Julia has an interface to produce a sequence of things (the iterator interface) and samplers should play nice with that as much as possible.