[d3d9] Several Wine device test fixes

[WinterSnowfall]

WIP to fix several low hanging d3d9-specific Wine device tests. Done.

So far:

• fixed SRCCOLOR2 & INVSCRCCOLOR2 caps being exposed with non-9Ex devices/interfaces (allegedly only WARP does this)
• fixed the test_refcount Wine test by preventing device child refcount underruns (we were already doing this in d3d8, thanks to Motor City Online)
• apply MultiplyTransform state block behavior universally (it is not d3d8 specific)
• operations on clip planes with an out of bounds index will reflect on the highest valid clip plane
• add surface format validations to D3D9Surface::GetDC() (note that valid formats still don't seem to work properly here, which might be something we'd want to revisit at some point)
• fixed a d3d8/d3d9 difference of expected error codes on failed device reset (due to losable resurces)
• fixed d3d9 state block clearing and recording behavior (the validations work similarly to the system we have in d3d8, but have to be implemented differently due to d3d9 state block implementation differences).

[WinterSnowfall]

Obligatory indication of progress:

Current master: 0020:device: 154593 tests executed (73 marked as todo, 0 as flaky, 1584 failures), 6 skipped.
This PR: 0020:device: 154308 tests executed (73 marked as todo, 0 as flaky, 845 failures), 6 skipped.

[WinterSnowfall]

Gave this PR a round of testing with ~100 d3d9 games and threw in some d3d8 games for good measure (although d3d8 is largely unaffected by these changes). Seems fine.