Remove explicit call to InfiniteMPS in VUMPS

[AFeuerpfeil]

I remove the explicit call to InfiniteMPS in VUMPS and instead use gaugefix! in the gauge_stetp!.

[lkdvos]

While I left a small comment on the implementation here, I think I would be happier with a more fundamental change where instead of the uniform gauge happening in the InfiniteMPS constructor, we actually use the dedicated function:
Something along the lines of:

gaugefix!(similar(it.state.mps), copy(it.state.AR), ...; kwargs...)

[lkdvos]

Suggested change

	ψ′ = typeof(mps)(it.state.mps.AR; alg_gauge.tol, alg_gauge.maxiter)
	ψ′ = typeof(mps)(copy(it.state.mps.AR); alg_gauge.tol, alg_gauge.maxiter)

Otherwise this will share the array with the input, which is probably not what we want here.
This being said, I think if you overload similar for your array type it keeps its type.

[AFeuerpfeil]

Thanks for the recommendation!

[github-actions]

Your PR no longer requires formatting changes. Thank you for your contribution!

[lkdvos]

Looks like the similar call is actually not sufficient as that does not allocate any of the tensors, I think copy is probably enough to fix this

[lkdvos]

Just looked at this a bit more - the IDMRG2 failures are most likely because the spaces of the MPS are changing, so my proposed solution in terms of an in-place function doesn't work that straightforwardly, since gaugefix! really does assume all tensors are the correct sizes. I'm not entirely sure if there is an easy way around this, as we were using the constructor as a trick to bring the MPS tensors back into a compatible form after the IDMRG which is allowed to change it...

[AFeuerpfeil]

@lkdvos: I don't see a good solution right now for IDMRG. For the sake of pushing this through to do some tests with VUMPS I have reverted the changes to IDMRG.

[lkdvos]

(OK for me, though it seems like you haven't pushed that 😉 )

[AFeuerpfeil]

I did push the IDMRG change, just messed up the commit message text with another commit. Therefore, it is called "format".
The test errors are due to VUMPS not converging for some reason. I do not really understand why.