ci(experimental): the GraalPy leg times out at 110 minutes after the shard split #2739

On main at ab7be2fb0, Experimental Tests run 41585968499 is the first uninterrupted GraalPy run since #2720 split the shards. It lost no workers, against eight before, so the heap problem is fixed. It timed out at 110 minutes instead (exit 124), with no test failures: pytest stopped "during the rest", after all 20 split shards had finished.

The shards ran slower than #2713's estimates. For example, tests/const took 502 s against an estimate of about 284 s, and the two tests/protocols/internet shards took 675 s and 817 s. Their test time adds up to about 6000 s before the rest begins, plus worker start-up for every run, against the step's 6600 s.

Fix it so the non-blocking leg stays green. Raise the step's timeout and the job's timeout-minutes, which GitHub-hosted runners allow up to 360. Or re-size the slowest shards. Or both. Use this run's JUnit seconds, since it is the first real measurement. Keep test_graalpy_shards_1963 and the #1963 note in workflows.rst in step.