[CI] Integration tests are all hitting OOM errors

[joerunde]

The last 2 PRs that went through the merge queue had their regression tests via /spyre-test fail.

See:

• tests: Remove xfails from previously-flaky fp8 tests #844 (comment)
• ci(test-matrix): let uv add --frozen tolerate a baked UV_NO_SYNC=1 #748 (comment)

Tests fail with this memory error then hang and hit timeouts:

  [unspecified]  ERRR 10.09.2026 22:30:06.974412 [          memory_allocator.cpp: 208] Out of memory. No free blocks available.
  [unspecified]  ERRR 10.09.2026 22:30:06.974516 [                  ras_base.hpp:  74] {"AllocSize":12288,"action":"information","category":"software","code":"0xf150","description":"Could not allocate memory for the calibration mapping","message":"Device memory allocation failed","name":"RAS::TSCALIBRATOR::DeviceMemAllocFailed","severity":"ERROR","step":"Open ticket","type":"runtime_error"}

The same failure is happening in runs triggered from torch-spyre as well, like this one: torch-spyre/torch-spyre#4429 (comment)

[joerunde]

Corroborated by test failures in this recent dependency bump PR as well: #851