test_spmv_bsrmatrix_controls_pattern incorrect epsilon?

[e10harvey]

In #1774, I noticed that the epsilon for this test is always using the float16 epsilon while it looks like this test runs for all types: float, double, etc. Shouldn't epsilon be:

static_cast<double>(
        std::numeric_limits<typename y_view_t::value_type>::epsilon());

Instead of

static_cast<double>(
        std::numeric_limits<Kokkos::Experimental::half_t>::epsilon());

in Test_sparse_spmv.hpp?

[e10harvey]

Relevant discussion thread: #1774 (comment)

[cwpearson]

The float16 tolerance should only be used when the algorithm under test is the tensor core algorithm. I wonder if this slipped through a refactor somehow.

[cwpearson]

This is resolved by removing the code in #1922