Transcoder capabilities discovery and validation

[cyberj0g]

What does this pull request do? Explain your changes. (required)
This PR is the last part of #2123 implementation. It will allow Orchestrator to send tasks to correct Transcoder according to requested capabilities (codecs). Also it adds capability test at startup for Nvidia GPUs (which may support only a subset of capabilities).

This PR includes changes of install_ffmpeg.sh discussed here. Corresponding LPMS PR needs to be merged first.

Specific updates (required)

• add capability test function which is run on node startup
• add logic to select transcoder according to capabilities in O-T mode
• add logic to validate capabilities in OT mode

How did you test each of these updates (required)
Ran tests

Does this pull request close any open issues?
#2123

Checklist:

• Read the contribution guide
• make runs successfully
• All tests in ./test.sh pass
• [n/a] README and other documentation updated
• [n/a] Pending changelog updated

Additional notes:
One missing piece is automatic capability (codec) detection from segment. Currently, if segment is encoded with codec, for which T doesn't support decoding - and decoding capability is not explicitly requested, it will bypass capability-based T selection logic and may fail, even if O has Ts which could decode it. Should go-livepeer has such segment codec check (at which point?), or it's best to handle on MistServer side?

[oscar-davids]

LGTM!

[cyberj0g]

Non-obvious quirk on multi-GPU system I just realized: there's no GPU-level segment scheduling logic, and capability discovery will provide a list of capabilities from less capable GPU, so if user has a rig with 5 x RTX 3090 and 1 GTX 745 to hook up the monitor, no codecs except H.264 would show up in capabilities, unless -nvidia argument excludes latter GPU. It worth documenting.

[cyberj0g]

Thanks for review, all comments addressed.

[darkdarkdragon]

LGTM after squash/rebase