turnip/ci: Bump the number of tests per group

Due to the slow startup time of deqp-vk, the previous default of 500 tests per group caused the jobs to run up to twice as slowly compared to using a higher number of tests per group.

I've attached a graph illustrating the relationship between tests_per_group and runtime. The graph is specifically for radv-raven-vkcts, but the trend is similar across all VKCTS runs.

While the long-term solution is to increase the default in deqp-runner for deqp-vk, this MR takes a more immediate approach by increasing the number of tests per group for all subsets of the deqp-runner suites for turnip, allowing for a reduction in the fractions. It’s worth noting that increasing tests_per_group only benefits VKCTS, GLCTS remains unaffected.

These changes enable us to either run more tests or decrease runtime, and in some cases, achieve both simultaneously. I've reduced the fractions for the merge jobs, while maintaining a runtime of around 10 minutes. The a660-vk job has been consistently taking too long, so I increased the fraction of the gmem run. However, if preferred, this can be adjusted by maintaining the fraction on the basic test set instead. We can now also complete a non-fractional run for the a618-vk-full job.

A notable highlight is the a630-vk-full job, which used to take ~130 minutes but now completes in about 60 minutes.