JIT: Improve inliner (constant feeds constant test, foldable calls)

[EgorBo]

Contribution to @tannergooding's work in #50675

static void Test()
{
    CheckHW();
}

static void CheckHW()
{
    if (!Avx2.IsSupported)
    {
        Console.WriteLine("No AVX2 :(");
    }
}

Current codegen for Test:

G_M24707_IG01:           
       4883EC28             sub      rsp, 40
G_M24707_IG02:             
       E8EF5CFEFF           call     CheckHW()  ;; <-- not inlined
       90                   nop
G_M24707_IG03:         
       4883C428             add      rsp, 40
       C3                   ret


; weight= 79 : state  40 [ call ]
; weight= 25 : state  45 [ brtrue.s ]
; weight= 66 : state 102 [ ldstr ]
; weight= 79 : state  40 [ call ]
; weight= 19 : state  42 [ ret ]
; 
; Inline candidate looks like a wrapper method.  Multiplier increased to 1.
; Inline candidate callsite is boring.  Multiplier increased to 2.3.
; calleeNativeSizeEstimate=268
; callsiteNativeSizeEstimate=55
; benefit multiplier=2.3
; threshold=126
; Native estimate for function size exceeds threshold for inlining 26.8 > 12.6 (multiplier = 2.3)

New codegen for Test:

G_M24707_IG01:           
G_M24707_IG02:             
       C3                   ret


; weight= 79 : state  40 [ call ]
; weight= 25 : state  45 [ brtrue.s ]
; weight= 66 : state 102 [ ldstr ]
; weight= 79 : state  40 [ call ]
; weight= 19 : state  42 [ ret ]
; 
; Inline candidate looks like a wrapper method.  Multiplier increased to 1.
; Inline candidate has an arg that feeds a constant test.  Multiplier increased to 2.
; Inline candidate has 1 call(s) which is guaranteed to be folded
; Inline candidate has const arg that feeds a conditional.  Multiplier increased to 5.
; Inline candidate callsite is boring.  Multiplier increased to 6.3.
; calleeNativeSizeEstimate=268
; callsiteNativeSizeEstimate=55
; benefit multiplier=6.3
; threshold=346
; Native estimate for function size is within threshold for inlining 26.8 <= 34.6 (multiplier = 6.3)

As far as I understand we're going to switch to "PGO-based policy" for inliner in future but maybe some of these pieces will be useful there too?

[tannergooding]

For some future PR, we could also track the intrinsic id here.

This would allow us to determine, at least the following interesting pieces:

• Number of intrinsic calls
• Number of hardware intrinsic calls
• Number of Type.op_Equality calls

There may be other interesting bits as well, but these are at least ones that have one or more of:

• significantly decreased cost (hwintrinsics are instructions not calls)
• additional optimizations (many intrinsics have CSE support)
• potential dead code elimination (type equality checks, particularly where T : struct)

[AndyAyersMS]

This is now going to require making String.Length a new-style intrinsic.

[EgorBo]

Will re-think this one in a new PR.