Random CI failures for Deterministic Device Reduce (RFA) with different policies

[srinivasyadav18]

Deterministic Device Reduce (RFA) tests fails randomly with different policies (block thread and items per thread combinations) on specific seeds : 839658673, 1540316222

Problem

The error check performed in the test catch2_test_device_reduce_deterministic.cu#L193 requires device result to be with in 1% of host reduce. Although both device results are exactly equal in catch2_test_device_reduce_deterministic.cu#L196, the host and device results are closer but far apart to be treated as error (i.e more than 1%).

Solution

Increase the tolerance of host and device result comparision to 5%.