[SYCL] Refactor check_device_code/vector LIT tests

[uditagarwal97]

This PR split check_device_code/vector LIT tests into two: one without preview-breaking-flag and one with the flag.
This would increase test coverage and also make it easier for us to switch tests during ABI-breaking window.

[uditagarwal97]

@intel/llvm-gatekeepers the code formatter job is failing in undef deprecator because this PR added a test where LLVM IR uses undef. See llvm/llvm-project#118506

Running: git diff -U0 --pickaxe-regex -S '([^a-zA-Z0-9#_-]undef[^a-zA-Z0-9_-]|UndefValue::get)' 47630fea32e28d7b6f93f824876608b7f084737d aa16eb44e70b2cf82bbd2ed9f52fe8825b3d8420

sycl/test/check_device_code/vector/vector_as_preview.cpp
sycl/test/check_device_code/vector/vector_bf16_builtins_preview.cpp
sycl/test/check_device_code/vector/vector_convert_bfloat_preview.cpp
sycl/test/check_device_code/vector/vector_math_ops_preview.cpp
sycl/test/check_device_code/vector/vector_as.cpp
sycl/test/check_device_code/vector/vector_bf16_builtins.cpp
sycl/test/check_device_code/vector/vector_convert_bfloat.cpp
sycl/test/check_device_code/vector/vector_math_ops.cpp
error: undef deprecator failed

Do we have a BKM on how to handle such failures? Just ignore them or open a GH issue for front-end/middle-end team?

[aelovikov-intel]

Do we have a BKM on how to handle such failures?

IMO, just merge.

[uditagarwal97]

Failure in E2E test (HostInteropTask/interop-task-cuda-buffer-migrate.cpp) on CUDA in unrelated and reported here #17026
Skipping Windows E2E job - which failed earlier due to an infrastructure issue - as this PR only modifies Linux-only LIT tests.