More XARCH PUTARG_STK CQ improvements

[SingleAccretion]

Included:

Dissolve structs into primitives for PUTARG_STK

The primary benefit of this change is "free" support for
complex addressing modes, which is always desirable when
we are loading a primitive (as opposed to the large struct
case, where we would not want to use the 3-operand LEA
in a loop, but instead cache the address in a register).

The additional (future) benefit is that we will no longer
need to mark the source local as DNER, once LCL_VAR sources
for struct PUTARG_STKs are supported.

And also:

Contain PUTARG_STK sources for "push [mem]"

The latter change is responsible for some pretty nice x86 diffs for TYP_FLOAT sources:

-       vmovss   xmm0, dword ptr [ebp+0CH]
-       sub      esp, 4
-       vmovss   dword ptr [esp], xmm0
-       vmovss   xmm0, dword ptr [ebp+08H]
-       sub      esp, 4
-       vmovss   dword ptr [esp], xmm0
-       vmovss   xmm0, dword ptr [ebp+10H]
-       sub      esp, 4
-       vmovss   dword ptr [esp], xmm0
+       push     dword ptr [ebp+0CH]
+       push     dword ptr [ebp+08H]
+       push     dword ptr [ebp+10H]

We are expecting good improvements, with a few regressions due to different RA choices.

There were too many x86 diffs for SPMI in CI, plus there is an SPMI outage, so here is the full locally obtained version: win-x86.

Highlights:

benchmarks.run.windows.x86:      -9.6K
libraries.crossgen2.windows.x86: -125K
libraries.pmi.windows.x86:       -55K

[ghost]

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

Issue Details

---

Included:

Dissolve structs into primitives for PUTARG_STK

The primary benefit of this change is "free" support for
complex addressing modes, which is always desirable when
we are loading a primitive is  (as opposed to the large
struct case, where we would not want to use the 3-operand
LEA in a loop, but instead cache the address in a register).

The additional (future) benefit is that we will no longer
need to mark the source local as DNER, once LCL_VAR sources
for struct PUTARG_STKs are supported.

And also:

Contain PUTARG_STK sources for "push [mem]"

The latter change is responsible for some pretty nice diffs for TYP_FLOAT sources:

-       vmovss   xmm0, dword ptr [ebp+0CH]
-       sub      esp, 4
-       vmovss   dword ptr [esp], xmm0
-       vmovss   xmm0, dword ptr [ebp+08H]
-       sub      esp, 4
-       vmovss   dword ptr [esp], xmm0
-       vmovss   xmm0, dword ptr [ebp+10H]
-       sub      esp, 4
-       vmovss   dword ptr [esp], xmm0
+       push     dword ptr [ebp+0CH]
+       push     dword ptr [ebp+08H]
+       push     dword ptr [ebp+10H]

Author:	SingleAccretion
Assignees:	-
Labels:	area-CodeGen-coreclr, community-contribution
Milestone:	-

[SingleAccretion]

@dotnet/jit-contrib

[BruceForstall]

The diffs look great! Thanks for the contribution!