crashes on Linux/NVidia from ctest --tests-regex Semaphore -j 20

Running `ctest --tests-regex Semaphore -j 20` causes many tests to segfault:
```
The following tests FAILED:
478 - VkLayerTest.ExternalTimelineSemaphore (SEGFAULT)
548 - VkLayerTest.QueueSubmitTimelineSemaphoreBadValue (SEGFAULT)
549 - VkLayerTest.QueueBindSparseTimelineSemaphoreBadValue (SEGFAULT)
551 - VkLayerTest.QueueSubmitBinarySemaphoreNotSignaled (SEGFAULT)
553 - VkLayerTest.InvalidExternalSemaphore (SEGFAULT)
557 - VkLayerTest.Sync2InvalidSignalSemaphoreValue (SEGFAULT)
558 - VkLayerTest.InvalidSemaphoreCounterType (SEGFAULT)
580 - VkLayerTest.InvalidVkSemaphoreTypeCreateInfoCore (SEGFAULT)
602 - VkLayerTest.QueueSubmitWaitingSameSemaphore (SEGFAULT)
1041 - VkLayerTest.SwapchainAcquireImageNoBinarySemaphore (SEGFAULT)
1042 - VkLayerTest.SwapchainAcquireImageNoBinarySemaphore2KHR (SEGFAULT)
1056 - VkLayerTest.SwapchainAcquireImageWithSignaledSemaphore (SEGFAULT)
1496 - VkPositiveLayerTest.LongSemaphoreChain (SEGFAULT)
```

These all crash because vkCreateDevice() returns VK_ERROR_INITIALIZATION_FAILED. I suspect we're violating some sort of resource limit in this driver. Linux / RADV and Windows / NVidia cannot reproduce this problem.

Start by reproducing `ctest --tests-regex Semaphore -j 20` on Linux/NVidia and inspect the listed semaphore tests where `vkCreateDevice()` returns `VK_ERROR_INITIALIZATION_FAILED`. Compare resource usage and behavior with Linux/RADV and Windows/NVidia, then determine whether the failures are caused by a driver resource limit. Done means the semaphore tests no longer segfault in the affected environment, or the limitation is documented and handled safely.

Written by the indexing model from the issue text.