r300: CI

This is still a draft, because while everything seems to work, there are still some problems:

• I still need the "when: always" rule hack in the final job to show up a all in the pipeline (I honestly don't understand the farm/r300 rules part I added at all, that was mostly copy pasted from other drivers looking at the commits doing similar CI enabling, so there is probably a lot of mistakes there).
• the B2C container is using a custom kernel with radeon module, need to upload it somewhere at fdo
• custom image: registry.freedesktop.org/mupuf/ci-tron/mesa-trigger:2024-01-05.1

However hopefully at this point it should be ready for some feedback.

Anyway, this MR introduces 3 r300 runners and adds a r300 dEQP gles2 job. This is intended for a manual runs. Regarding the used HW, there are 3 ancient DELL optiplex 755 workstations with even more ancient RV530 GPUs. This is setup behind a RPi4B gateway running ci-tron.

Some notes:

• this is running in my closet with all the usual caveats of rural countryside in Czechia (slow internet connection 30/3Mbps + not so reliable power network)
• no UPS, so in the case of power outage this all goes down (maybe once a month on average, mostly in winter/summer during storms). I could in theory get one UPS so that it survives for a few minutes which should be enough for most of the power outages and than could reduce the shutdowns to maybe two longer events per year
• related to it, when the power goes up again, the gateway will try to netboot but at that point the internet connection is not up yet and the boot will thus fail, and needs to be restarted manually, this could be probably fixed at the gateway?
• the runner bring-up is SLOW. It takes like 100 seconds after the job starts before the runner is even powered on by the gateway and than around 3 more minuts before the actual testing starts. The dEQP itself is actually quite fast, with runtime below 3 minutes. So like 6-7 minutes in total which is not so bad, but most of it is not the actual testing. :-(
• one caveat of the machines is that the board only allows 25W for the PCIe which is not enough for the cards, so I had to use PCIe risers with external power connectors, atm I have only x1 to x16 so the cards are just using PCIe x1 even though the boards have x16, I have x16-x16 risers ordered, however the reduced bandwidth seems to not matter much for the dEQP speed at least.
• the test reliability itself looks good, no flakes so far using the bin/ci/ci_run_n_monitor.sh script for stress testing

Huge thanks to: @dh for arranging for me the DELL optiplex 755 workstations and the risers for free @mupuf for unbelievable support setting this all up

The plan is to follow up with more tests later (piglit, traces, testing the draw paths, and maybe also some nine testing).