avoid twist allocating in rrules when not required

[lkdvos]

This is a temporary non-breaking solution before #305 fully resolves the issue, to already include in a patch release.

@leburgel tagging you here

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.

Files with missing lines	Coverage Δ
ext/TensorKitChainRulesCoreExt/linalg.jl	100.00% <100.00%> (ø)
ext/TensorKitChainRulesCoreExt/tensoroperations.jl	93.13% <100.00%> (+0.20%)	⬆️

🚀 New features to boost your workflow:

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

[leburgel]

I think there's two more in the rrule(::typeof(⊗), ...) in ext/TensorKitChainRulesCoreExt/linalg.jl. I assume these will have the same issue, so might be good to change these here too?

[lkdvos]

Unfortunately, the non-passing tests seem to be unrelated. I still haven't figured out a good way to make testing truncated SVD compatible with finite-difference checks, since this more or less requires that the variations are small enough that the truncation selects the same subspace every time.
I think the only way around this is to carefully craft input matrices with well-defined spectra, which can and should be done in a separate PR