[d3d8] GetInfo rework & a couple of CopyRects validation tweaks

[WinterSnowfall]

I've reworked our implementation of GetInfo() to more closely follow what d9vk is doing, namely:

• pQuery->Issue() will never fail in d9vk (and even if can fail according to spec, we shouldn't care much, as pQuery->GetData() would also then fail).
• D3DDEVINFOID_RESOURCEMANAGER and D3DDEVINFOID_VERTEXSTATS, though technically supported by d3d8 are not implemented in d9vk. D3D9's CreateQuery() will always return D3DERR_NOTAVAILABLE for both as it stands, thus we shouldn't really speculate on the outcome (removed the "unlikely" hint).
• According to specs, D3D9 GetData() should always be called with D3DGETDATA_FLUSH for the queries that D3D8 supports.
• D3D9 GetData() returns are inline with what GetInfo() should return if it reaches that point.

The above allows us to get rid of the goto and restructure the flow a bit.

Also as part of this PR I've changed some CopyRects behavior:

• removed format difference check from copyTextureBuffers, since this is now done globally on CopyRects calls, as per spec.
• CopyRects depth stencil checks are done after the format difference checks, so there's no need to check both src and dest.

[WinterSnowfall]

Fixed a couple of other things I stumbled into:

• We were creating an invalid auto depth stencil in cases where the D3D9 device was created with the EnableAutoDepthStencil present param flag set to FALSE. The stencil surface would cause a crash in these cases on any direct calls, such as GetDesc(). Keeping it as nullptr is pretty much the correct behavior here, since if the application doesn't set a depth stencil, the runtime doesn't generate one.
• I honestly don't see how it's possible to create a depth stencil with D3DFMT_UNKNOWN as everything validates that away, however using D32 was incorrect, because it's technically unsupported. Replaced it with D24X8, although I don't think it's actually going to end up used anyway.