fpltool extract: parallel video decoding, optional hw (VA-API) decoding

[flipflip8952]

fpltool:

• Decode video in parallel when extracting camera data to a ROS bag. Each group of pictures (GOP) is decoded by its own decoder in its own thread. All messages go through a queue and are written in the order they are in the log.
• Add -j, --jobs option for the number of parallel video decoders (default: number of CPUs - 2, at least 1)
• Add -a, --accel option to select the hw acceleration of the video decoders: SW, HW or AUTO (default: SW)
• Write the last frame of each camera, which was decoded but not written
• Count errors, so that the limit of 100 errors has an effect

fpsdk_common:

• video: make hardware (VA-API) decoding work. The filter graph is set up from the format of the first decoded frame. Hardware frames are scaled on the GPU, downloaded as yuv420p and converted to the output pixel format in software. HwAccel::HW fails if the decoder cannot use the hardware, HwAccel::AUTO falls back to software decoding.
• video: document hardware vs. software decoding performance in video.hpp
• fpl: fix rec_time in the CamData info string

Docker, devcontainer:

• Build FFmpeg with the h264 and hevc VA-API hwaccels
• Pass /dev/dri to the trixie-dev container
• Install the time utility
• Move persistent devcontainer data to .devcontainer/.persist
• Add .devcontainer/helper.sh script to help setting things up