Make TT device tests faster in CI

[matejcik]

The multi-core PR #2462 helps cut down run-time of tests on the local machine. The same trick won't work in CI however, as every runner only has one CPU so there is nothing to parallelize within a single job.

Instead, we could split up the device tests into multiple jobs: one per coin and one for all the rest would be a reasonable start IMO. This would require the ability to merge UI test results from multiple separate runs for the purpose of --ui-check-missing and generating the final report, which is also needed for #2493

[grdddj]

If we succeed with this, it can also be applied to Connect tests, which currently take approximately the same time as device/UI tests.

Not with a big priority, as there are no jobs depending on Connect, but there are posttest and deploy jobs depending on device/UI tests. That however brings the question, if we split it into multiple jobs, can we then make a job that will only start after all of them finish?