ci: Rebalance LAVA jobs - Round 2

Following the deployment of the DUT Time per GitLab Job dashboard, we have identified several jobs for AMD, Intel, and Freedreno that are exceeding our target execution times.

This MR will continue the efforts from !31414 (merged)

Summary of changes

Based on this dashboard snapshot, job duration times and pending times, the rebalancing goes that way:

Affected Job/Component	Change Description	Reason	Impact	Current Job Duration (average in minutes)
Turnip (kingoftown jobs)	Decremented parallel jobs from 13 to 12 and increased dEQP fraction for a618_vk	Match number of available devices in Collabora's farm	Reduced job waiting time, optimized resource usage.	16.93
Freedreno (limozeen jobs)	Reduced parallel a618_gl job from 2 to 1, increased dEQP fraction from 2 to 5	Overload of available DUTs (6 DUTs, 7 jobs), causing delays	Balanced load, reduced pending times for DUTs, optimized job completion within 10 minutes.	14.11
Panfrost	Increased panfrost-g52-gl parallel jobs from 3 to 4	Job takes 14 minutes on average with rare retries, spare DUT available	Reduced job time to reach 10-minute mark.	14.53
AMD (radv-raven-traces)	Reassigned radv-raven-traces from zork to dalboz and added a merge-skips list	Zork devices overutilized with additional radeonsi-raven job	Reduced load on zork, balanced distribution of resources.	12.4
Radv (stoney-vkcts, stoney-angle)	Increased DEQP_FRACTION for radv-stoney-vkcts from 15 to 18, and for stoney-angle from 2 to 3	~95% DUT availability with some infra retries, requesting higher DEQP fraction.	Reduced test time, better handling of infra retries while utilizing available DUT capacity effectively.	14.4 and 14
Zink (anv-adl, anv-tgl skips)	Added skip tests for zink-anv-(adl|tgl) due to long-running tests	Many (10) tests take more than 1 minute to complete.	Excluded slow tests from merge pipelines to optimize execution time in affected jobs.	18.35 (adl) 17.44 (tgl)