Merge random pipeline generator

[TH3CHARLie]

This PR attempts to merge the random pipeline generator that is used to generate autoscheduling training data into main, originally from: https://github.com/halide/Halide/tree/abadams/random_pipelines/

One thing I'm not sure about is do we want to include

1. A similar driver code process.cpp like other apps that just test if the generator code can compile and run (see examples like bgu)
2. the autotune script that actually uses the generator to generate the schedule (see Andrew's original branch)

cc @abadams @alexreinking

[abadams]

I think we need to have some way to compile and run it, and some build target that does so. Maybe RunGen-based?

[alexreinking]

I have some feedback re: the handling of randomness in this. There are good utilities in C++ for generating uniform randomness. Using mod introduces bias. I'm not sure about the division thing for floats either.

[alexreinking]

This could at least take a const reference, but it could be more simply written as

vector<Expr> arguments{vars.begin(), vars.end()};

wherever it is used.

[alexreinking]

This is janky. We should really use the proper RNG types...

class Random {
    std::mt19937 eng_;
    uint32_t seed_;

   public:
    explicit Random(uint32_t seed) : eng_(seed), seed_(seed) {}

    void reseed(uint32_t seed) {
        eng_.seed(seed);
        seed_ = seed;
    }

    uint32_t seed() const { return seed_; }

    int integer(int min, int max) {
        std::uniform_int_distribution<int> dist(min, max);
        return dist(eng_);
    }

    bool boolean(double p = 0.5) {
        std::bernoulli_distribution dist(p);
        return dist(eng_);
    }

    float real() {
        std::uniform_real_distribution<float> dist(0.0f, 1.0f);
        return dist(eng_);
    }

    template <typename T>
    decltype(auto) element(T&& t) {
        auto n = std::size(t);
        std::uniform_int_distribution<std::size_t> dist(0, n - 1);
        return t[dist(eng_)];
    }
};

Random R;

Something like this should work.

[alexreinking]

No need for the _count if we use the element helper above.

[alexreinking]

Suggested change

	int op = rng() % comp_bin_op_count;
	return make_comp_bin_op[op](a, b);
	return R.element(make_comp_bin_op)(a, b);

[alexreinking]

No need for counts

[alexreinking]

Suggested change

	return cast(t, rand_float());
	return cast(t, R.real());

[alexreinking]

Suggested change

	rng.seed((int)seed);
	R.reseed(seed);

[alexreinking]

Suggested change

	GeneratorParam<int> seed{"seed", 1};
	GeneratorParam<uint32_t> seed{"seed", 1};

[alexreinking]

Newline

[alexreinking]

Please insert this in alphabetical order (and fix HelloBaremetal, apparently)