We don't want our gcp buckets to get overfilled with builds from
external devs.

Not all fields will be identical segment-to-segment,
particularly DurationSecs. This caused unnecessary
transcoder re-initialization.

* Update LPMS to latest.

* protoc: Regenerate

* Clean up inactive sessions

Not cleaning up the sessions causes quite a big memory leak which in effect causes OOMKilled

* Clean up inactive sessions

Not cleaning up the sessions causes quite a big memory leak which in effect causes OOMKilled

…cy score (#3086)

Problem: We never cleaned the known sessions, even if all of them had too low latency score to use; example of what sometimes happens:
1) Broadcaster accumulated 8 known sessions, but all had too low latency score
2) Session refresh was not triggered because we had a lot of knows session in the pool
3) At the same time, we didn't have sessions in unknown session pool
4) Selection checked that it cannot use the known session and tried to select from unknown sessions (but the pool was empty or there were no sessions fulfilling the condition of max price or perf score)

Co-authored-by: Thom Shutt <[email protected]>

This commit introduces a mechanism for gateways to configure the timeout during the discovery process. It adds the `DiscoveryTimeout` CLI argument, with a default value of 500ms, consistent with the previous codebase.