Remove BangBang.possible

[sunxd3]

In light of JuliaFolds2/BangBang.jl@e679bae, we can removed the code as they are implemented in BangBang.

The changes has not been released, so should verify afterward. And this PR should not be merged right now.

related: fix #530

[sunxd3]

I don't understand why Document-CI is failing, FLoop is not in BangBang's deps.

[yebai]

I don't understand why Document-CI is failing, FLoop is not in BangBang's deps.

Maybe Floop is in the deps on another package? I think Julia's package manager has a mechanism to figure which package requires a specific deps.

[devmotion]

I don't understand why Document-CI is failing, FLoop is not in BangBang's deps.

MLUtils is a docs dependency and depends on FLoops - however FLoops does not support BangBang >= 0.4 currently.

[devmotion]

MLUtils is used in a single example:

DynamicPPL.jl/docs/src/tutorials/prob-interface.md

Lines 112 to 141 in abcf584

	using MLUtils
	function cross_val(
	dataset::AbstractVector{<:Real};
	nfolds::Int=5,
	nsamples::Int=1_000,
	rng::Random.AbstractRNG=Random.default_rng(),
	)
	# Initialize `loss` in a way such that the loop below does not change its type
	model = gdemo(1) | (x=[first(dataset)],)
	loss = zero(logjoint(model, rand(rng, model)))
	for (train, validation) in kfolds(dataset, nfolds)
	# First, we train the model on the training set, i.e., we obtain samples from the posterior.
	# For normally-distributed data, the posterior can be computed in closed form.
	# For general models, however, typically samples will be generated using MCMC with Turing.
	posterior = Normal(mean(train), 1)
	samples = rand(rng, posterior, nsamples)
	# Evaluation on the validation set.
	validation_model = gdemo(length(validation)) | (x=validation,)
	loss += sum(samples) do sample
	logjoint(validation_model, (μ=sample,))
	end
	end
	return loss
	end
	cross_val(dataset)

Since it's a quite heavy dependency, I wonder if we could avoid it.

[sunxd3]

Okay, the situation is a bit troublesome.
BangBang v0.4 moved to Accesssors, so all the Setfield related code are erroring out.
Now, either we led some effort in backporting the possible fix to v0.3, or ourselves move to Accessors too.
The interface diff between Setfield and Accessors should be minimal for our purpose, though.

[yebai]

Now, either we led some effort in backporting the possible fix to v0.3, or ourselves move to Accessors too.

Let's move to Accessors too. @sunxd3, can you create a PR for that next week?

[coveralls]

Pull Request Test Coverage Report for Build 8745141284

Details

• 0 of 0 changed or added relevant lines in 0 files are covered.
• No unchanged relevant lines lost coverage.
• Overall coverage decreased (-0.03%) to 81.821%

---

Totals
Change from base Build 8745047931:	-0.03%
Covered Lines:	2615
Relevant Lines:	3196

---

💛 - Coveralls