Propagate typeof() during inlining

[EgorBo]

Closes #40381

Example:

static void Main(string[] args)
{
    Console.WriteLine(Foo(typeof(float)));
}

static bool Foo(Type t)
{
    if (t == typeof(int))
        Console.WriteLine("1");
    if (t == typeof(uint))
        Console.WriteLine("2");
    if (t == typeof(long))
        Console.WriteLine("3");
    if (t == typeof(ulong))
        Console.WriteLine("4");
    return t.IsValueType;
}

Codegen for Main():

       4883EC28             sub      rsp, 40
						;; size=4 bbWeight=1    PerfScore 0.25
G_M60419_IG02:              ;; offset=0004H
       B901000000           mov      ecx, 1
       FF15B1F43800         call     [System.Console:WriteLine(bool)]
       90                   nop      
						;; size=12 bbWeight=1    PerfScore 3.50
G_M60419_IG03:              ;; offset=0010H
       4883C428             add      rsp, 40
       C3                   ret  

It now:

1. Inliner now understands that typeof() at that call-site is a constant arg -> recognizes 4 foldabale branches in this inlinee:

Inline candidate has 4 foldable branches.  Multiplier increased to 8.

2. Propagates typeof() by value during inlining - it helps to eliminate those branches and fix RyuJIT: IsValueType and IsAssignableFrom optimizations aren't inlining friendly. #40381

Regressions are due to inlining, RA and some cases where CSE/Hoisting doesn't "undo" propagation, e.g.: https://www.diffchecker.com/ukNlUweZ

Also a lot of regressions coming from the fact that we never do CSE across blocks, e.g.:

static void Bar(int a) => Foo(a, typeof(int));

static void Foo(int a, Type t)
{
    if (a == 42)
        Console.WriteLine(t);
    else
        Console.WriteLine(t);
}

codegen diff for Bar: https://www.diffchecker.com/mWkBicM0
but it's not really a terrible perf regression I'd guess...

[ghost]

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

Issue Details

---

Quick experiment, locally saw more improvements than regressions from it

Author:	EgorBo
Assignees:	-
Labels:	area-CodeGen-coreclr
Milestone:	-

[EgorBo]

PTAL @dotnet/jit-contrib @AndyAyersMS
overall, good diffs. Size regressions mostly due to the fact we propagate typeof() into different basic blocks and then don't CSE it back - but it might be a perf improvement actually, because we don't call the helper when it's not needed.

[EgorBo]

@AndyAyersMS anything else?