SemaHLSL's FlattenedTypeIterator does not handle bit fields properly.

There are a variety of repros possible, such as: casting from one struct to another where one has bitfields and the other has a series of uints equivalent to the size of the bitfield storage, or from a literal int 0 to a struct with a starting bitfield of enum type.

Example 1:

// RUN: %dxc -T lib_6_6 -HV 2021 %s

struct SomeStructWithBitfields {
    uint32_t m1 : 8;
    uint32_t m2 : 16;
    uint32_t m3 : 6;
};

struct StructWithUint {
    uint32_t u;
};

StructWithUint cStructWithUint;

export uint32_t SomeFuncCastingStructs()
{
    // error: cannot convert from 'const StructWithUint' to 'SomeStructWithBitfields'
    SomeStructWithBitfields bf = (SomeStructWithBitfields)cStructWithUint;
    return bf.m2;
}

Example 2:
This one may have a different root cause, or involve an additional bug, which should be broken out into another issue during investigation if so.

// RUN: %dxc -T lib_6_6 -HV 2021 %s

enum SomeEnum { Val1 };

struct SomeStructWithEnum
{
    // Uncommenting the uint32_t field makes the cast ok somehow...
    //uint32_t m1 : 16;
    SomeEnum m3 : 3;
};

export int SomeFuncUsingEnum()
{
    // This cast only succeeds when the first bitfield is not an enum.
    // Uncommenting the uint32_t field gets past taht, but then there's a
    // crash due to issue #5257
    SomeStructWithEnums = (SomeStructWithEnum)0;
    s.m3 = Val1;
    return (int)s.m3;
}

Example 3 (expect some diagnostics casting to uint here):

// RUN: %dxc -T lib_6_6 -HV 2021 %s

struct SomeStruct2
{
    uint32_t m1 : 16;
    uint32_t m2 : 19;
    uint32_t m3 : 3;
};

export uint SomeFunc2()
{
    SomeStruct2 s = (SomeStruct2)0;
    // Expect some error or warning when casting to uint, since SomeStruct2 is larger than one uint:
    return (uint)s;
}

Reproduce the three HLSL cases with the shown dxc -T lib_6_6 -HV 2021 commands and inspect SemaHLSL's FlattenedTypeIterator while comparing bitfield and enum handling. Determine whether the enum case is separate from issue #5257, then add regression coverage for the confirmed casts and expected diagnostics.