…hw decode

Two bugs in the D3D11VA seamless transition path:

1. Black flash (RendererDX11.cpp). The NV12 YUV shader variants were
   getting their own blend uniforms (`weights` for cubic, `delta` for
   linear) created alongside the RGBA variants' uniforms. Those YUV
   uniforms start dirty with zero-initialized data (base CShaderUniform
   ctor sets m_bDirty=true). The first time a freshly-created YUV
   variant's Apply() runs, it uploads zeros to slot b1 — clobbering the
   correct weights the RGBA variant's Apply just bound there — and the
   cubic blend produces (0,0,0) for one frame. After that the uniform
   is clean and never rebinds b1, which is why only the very first
   cubic draw of each new clip rendered black. Invisible at app startup
   because fade-in composites over it; visible at every seamless swap
   where alpha jumps straight to 1.

   Fix: don't create uniforms on the YUV variant. It's only ever bound
   mid-draw via the shader swap in DrawTexturedQuad, immediately after
   the RGBA variant's Apply has already bound the correct CB at b1.

2. Forward jump (Player.cpp). While the next clip is waiting for a
   seamless swap it isn't rendered, but we were still Update()ing it
   every tick — so its decoder queue drained as fast as playback, and
   by swap time the first frame popped was N seconds into the clip
   instead of frame 0. With hw decode this showed up as a visible
   forward jump; sw decode hid it because the decoder couldn't keep up.

   Fix: gate the nextClip Update() on the pending transition type. The
   decoder thread keeps running and fills its queue to the backpressure
   cap regardless, so the queue is fully buffered at swap time ready to
   pop frame 0.

Co-Authored-By: Claude Opus 4.7 (1M context) <[email protected]>