Add NVENC AV1 hardware video encoding for WebRTC

[larsers]

Add GPU-accelerated video encoding via NVIDIA NVENC to the WebRTC streamer, with a plugin/provider architecture that supports multiple codecs and falls back gracefully to software VP8 when no NVIDIA GPU is available.

Key changes:

Provider/registry framework:

• Add EncoderProvider/DecoderProvider abstract interfaces with singleton registries and priority-based selection
• Add CompositeEncoderFactory/CompositeDecoderFactory that query all registered providers at runtime
• Replace hardcoded VP8-only encoder factory with composite factory

NVENC encoding:

• Add codec-agnostic NvencVideoEncoder driven by NvencEncoderConfig (codec GUIDs, bitrate limits, per-codec callback hooks)
• Add AV1 NVENC provider with runtime GPU capability detection
• Add CudaContext singleton for CUDA device management

Native ARGB/ABGR frame path:

• Add AbgrBuffer and CvdAbgrVideoFrameBuffer for packed frame formats, eliminating CPU-side I420 conversion when NVENC is used
• Modify DisplayHandler to pass native ARGB/ABGR frames directly
• Modify VideoTrackSourceImpl for ABGR routing and monotonic timestamp enforcement

Build system:

• Add rules_cuda, nv_codec_headers to MODULE.bazel
• Add override_dh_shlibdeps for CUDA/NVENC shared libs
• Use alwayslink = True for static initializer registration