Analyze reaching PHI in GetRangeFromAssertions (more branch foldings)

[EgorBo]

Some interesting diffs are produced.

I slowly mimic what SSA-based GetRange does, but purely on top of VNs and assertions. This should be, in theory, more powerful, doesn't require a separate "does overflow" path, can be called from Global Assertion Prop general optimizations.

simple example:

void Test(bool cond, int[] arr)
{
    int x = cond ? -1 : -2;   // [-2..-1]
    int y = arr?.Length ?? 0; // [0..Array.MaxLength]

    if (x > y) // always false -- it's now folded
        Console.WriteLine("below zero");
}

[dotnet-policy-service]

Tagging subscribers to this area: @JulieLeeMSFT, @dotnet/jit-contrib
See info in area-owners.md if you want to be subscribed.

[BoyBaykiller]

With this PR is i recognized as non negative?

for (int i = 0; i < arr.Length; i++)
{
   Console.WriteLine(i >= 0); // fold to true?
}

[EgorBo]

With this PR is i recognized as non negative?

for (int i = 0; i < arr.Length; i++)
{
   Console.WriteLine(i >= 0); // fold to true?
}

Unfortunately, still no. We can't encode PHI in VN for this case, so i basically has no VN inside the loop body in this case. SSA+Assertions and/or IV is the only way for loops.