Buildbot LoadCI integration

[Sonicadvance1]

LoadCI integration is a time consuming process. List of steps we need to support.

Step 1

• [ ] Switch CI system over to buildbot
  • [ ] Show buildbot steps in in Github view
  • [ ] Block merge when buildbot not ran (When not authenticated)
  • [ ] Allow Admin CI running via comment
  • [ ] Show results of LoadCI in comment
  • [ ] Block merging when performance has degraded
  • [ ] Allow manual override (Only authenticated users)
• [ ] Github event notification
  • [ ] Authentication check with buildbot group
  • [ ] LoadCI message stating changed performance on
  • [ ] Only merge LoadCI results to Postgres on main master merge

Step 2

• [ ] Discord integration
  • [ ] Show failures against main branch in bots channel
  • [ ] Kick off PR build directly from bots channel (Only authenticated users)

Step 3

• [ ] Load balancing / Job splitting
  • [ ] Every builder must build its own instance of FEX
  • [ ] Create runner families that should closely match performance to one another
  • [ ] Split LoadCI jobs to runner families

https://docs.google.com/document/d/1Km9ZFdSx932_z61kbV9DVkdJAg-OUqx-37KrsJmpDI8/edit#

[Sonicadvance1]

LoadCI programs

• dav1d - Ensures vector performance
• bytemark - Ensures a bunch of things including integer perf

[skmp]

@Sonicadvance1 run into this while looking for something else, I can contribute the C630 and the SD865 HDK into the CI pool as they have been replaced by the MBP M1 at this point.

They barely have enough ram for modern development, but should do just fine for LoadCI.

Plus, I have space for them to run 24/7 in the new office.

[Sonicadvance1]

We will want loadCI that can benchmark video decoding of formats and libraries that we care about. gstreamer, libva, h.264, h.265, VP8, VP9, AV1. There is a good chance we will fallback to software decoding in a /lot/ of cases so can't rely on hardware video decode.