Pass information through to `step` / `step_warmup` about whether a step will be discarded

[penelopeysm]

There are two situations in which we call step / step_warmup, and then proceed to throw away the transition object and only use the state:

1. initial discarded steps (burn-in)
2. thinning

AbstractMCMC.jl/src/sample.jl

Lines 201 to 208 in 390012e

	# Discard initial samples.
	for j in 1:discard_initial
	# Obtain the next sample and state.
	sample, state = if j ≤ num_warmup
	step_warmup(rng, model, sampler, state; num_warmup, kwargs...)
	else
	step(rng, model, sampler, state; kwargs...)
	end

AbstractMCMC.jl/src/sample.jl

Lines 229 to 235 in 390012e

	for _ in 1:(thinning - 1)
	# Obtain the next sample and state.
	sample, state = if i ≤ keep_from_warmup
	step_warmup(rng, model, sampler, state; num_warmup, kwargs...)
	else
	step(rng, model, sampler, state; kwargs...)
	end

We should pass an extra keyword argument into step and step_warmup to let implementations know whether they need to return a meaningful transition or not. This could have substantial performance benefits e.g. for Turing, where calculating the transition object requires model re-evaluation.

[penelopeysm]

This would also provide a mechanism by which TuringLang/Turing.jl#2639 could be fixed.