Instantiate broadcast for pullbacks

[kshyatt]

This prevents an axes check from exploding further up the stack when PEPSKit tries to call this

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.

Files with missing lines	Coverage Δ
src/pullbacks/eig.jl	92.72% <100.00%> (ø)
src/pullbacks/eigh.jl	86.13% <100.00%> (ø)
src/pullbacks/svd.jl	94.32% <100.00%> (ø)

🚀 New features to boost your workflow:

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

[Jutho]

The broadcasted approach was suggested by @lkdvos a while ago as a way to avoid an extra copy/allocation. If we then end up instantiating it again, this seems like a convoluted way to replace the simpler code that was there before Lukas’ suggestion, but I might be misremembering.

[kshyatt]

It's again our frenemy the GPU libraries not being able to "detect" that a Broadcasted across complicated GPU objects is itself spiritually a GPU object. If we can find what the previous implementation was (git blame, I guess), I can return to the drawing board and try to find a nicer way to achieve the same ends.

[lkdvos]

I don't think instantiate here means the same as materialize, or at least that's what I convinced myself of when I was looking at this, and I don't think this is materialized