First, the code in fgAddFieldSeqForZeroOffset that was adding zero-offset sequence to LCL_FLD directly was incorrect, because fgAddFieldSeqForZeroOffset adds a sequence for the address that the passed-in node represents, while adding a field sequence to LCL_FLD directly means adding it to the "implicit" LCL_FLD_ADDR it indirects. One can reproduce this issue via code like the following (with JitNoStructPromotion=1):

private static int Problem(StructWithRawFldPtr p)
{
    return ((StructWithIndex*)p.RawFldPtr)->Index;
}

struct StructWithRawFldPtr
{
    public nint RawFldPtr;
}

struct StructWithIndex
{
    public int Index;
    public int Value;
}

LCL_FLD   long   V00 arg0         u:1[+0] Fseq[RawFldPtr, Index] (last use)

The first two commits fix this issue, in fgAddFieldSeqForZeroOffset and similar code in ChangeOper and we now get the lovely combination of LCL_FLD + zero-offset sequence in the map.

LCL_FLD   long   V00 arg0         u:1[+0] Fseq[RawFldPtr] (last use) Zero Fseq[Index]

That opens a possibility for another bug of course: partial sequences, that is cases where we will have ADD((LCL_FLD PtrToStatic [Zero FldSeq]), OFFSET FldSeq) and fail to incorporate the Zero FldSeq into the PtrToStatic's VN. This is fixed in the third commit.

Part of #58312.
Contributes to #63768.

No diffs as expected.