Set default tol_max in gauging eigsolver

[leburgel]

Sets the default maximal tolerance in the solver used in the gauging eigsolve steps. I was having stability issues, where for a slowly converging orthogonalization the eigsolve step used a tolerance so high that it was messing up convergence altogether since the error kept increasing after each inaccurate eigsolve.

I don't have a minimal working example, but I think restricting the tol_max here can't hurt either way. Setting it at Inf seems a bit dangerous...

[codecov]

Codecov Report

All modified and coverable lines are covered by tests ✅

Files with missing lines	Coverage Δ
src/states/ortho.jl	86.56% <ø> (ø)

... and 2 files with indirect coverage changes

[lkdvos]

Yeah, that's probably way more sensible.

I'm okay with merging this as is, but if you have the time it would be even better to actually move this to the Defaults section.

[leburgel]

I'm okay with merging this as is, but if you have the time it would be even better to actually move this to the Defaults section.

I could just move the _GAUGE_ALG_EIGSOLVE definition to the Defaults module, rename it and then call that. But I think it's actually easier to just call Defaults.alg_eigsolve with the appropriate kwargs. I'll give that a go.

[leburgel]

I think it makes sense to still define a constant within ortho.jl but use Defaults.alg_eigsolve to construct it. In the end the only non-default part was ishermitian=false and tol_factor=1, so it's a bit cleaner.

It didn't make sense to me to move this specific definition to Defaults and then call it from there. I don't think this would be cleaner, and I don't see a more general use case for these specific settings.