Test truncated methods with GPU arrays

[kshyatt]

Try to address #138

[lkdvos]

I would slightly prefer to have these changes as specializations in the GPU extensions, but definitely happy to try them out like this

[kshyatt]

OK, got the column permutation for V in eigh_full for Diagonal working (what a mouthful...). Base.permutecols!! isn't GPU friendly as is.

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.

Files with missing lines	Coverage Δ
src/implementations/eigh.jl	94.80% <100.00%> (+0.36%)	⬆️
src/implementations/truncation.jl	90.80% <100.00%> (-0.50%)	⬇️

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.

[kshyatt]

We can probably use the accumulate approach by defining a custom struct such that:

struct Truncerr_BinaryAdder{T}
    p::T
end
(a::Truncerr_BinaryAdder)(x, y) = x + abs(y)^a.p

Then in the GPU extensions do:
GPUArrays.neutral_element(::Truncerr_BinaryAdder, T) = abs(zero(T))
and finally
accumulate(Truncerr_BinaryAdder(p), reverse(view(values, I); init = abs(zero(eltype(values)))

But that can happen in a separate PR since this is holding up TK.