multi-frame textured point cloud streaming

[bmegli]

The last step in hardware-video-streaming#5.

This needs only NHVD wrapper changes, adjusting function calls and initializing 2 hardware decoders.

Everything else happens in NHVD under the hood.

[bmegli]

Some subjective impressions after initial tests:

• works much better than P010LE + UV hack
• the required bitrate is much lower (e.g. 4 Mbit + 1 Mbit gives good results)
• uses more GPU (mainly on encoding side), with encoding and decoding on the same machine and max quality effort forced in VAAPI GPU usage exceeds 50%

[bmegli]

Before merging this, review interop.

To initialize multiple hardware decoders we pass:

• array of structures of hardware configurations
• with some integers and C style ASCII strings

Confront current implmementation with:

Microsft marshaling-different-types-of-arrays

[bmegli]

Looks ok.

In fact we have two interfaces implemented in NHVD .Net wrapper for convenience:

• the one that passes array of structs, as above, plus depth config
• simplified that takes network config struct and hardware config struct (for video streaming)

Down in C they both tap to the same function that takes pointer to hardware config (which can be both single element and array in C)

Some more docs will be needed after reworking NHVD/UNHVD later.