BDD MLE, modular compilation, alg. data types, M1 compatibility

[rtjoa]

Nearly all of the LOC changed are due to added examples. To see how the backend changed, start with the explanations of includes at the bottom of src/inference/inference.jl.

BDD MLE

See examples/qc/tour_2_learning.jl.

Modular compilation

The separation of WMC, BDD compilation, and pr supports BDD MLE. For an example that provides more general motivation, see the following function, which efficiently finds the exact max value of a DistUInt:

function maxvalue(x::DistUInt{W}) where W
    c = BDDCompiler(x.bits)
    v = 0
    ctx = true
    for i = 1:W
        bit, bitvalue = x.bits[i], 2^(W-i)
        if issat(c.mgr, compile(c, ctx & bit))
            ctx &= bit
            v += bitvalue
        end
    end
    v
end

Algebraic data types

See src/dist/inductive/list.jl for an example definition of an algebraic datatypes and test/dist/inductive/list_test.jl for its usage.

M1 compatibility

I ripped out TikzGraphs and TikzPictures (which depend on Tectonic, see this issue), and added a pointer to rtjoa/CUDD.jl:m1compat in the README.