Truncation composition

[mtfishman]

This adds support for composing truncation strategies, which is then used to enable mixing truncating with rtol, atol, and maxrank.

[codecov]

Codecov Report

Attention: Patch coverage is 50.00000% with 10 lines in your changes missing coverage. Please review.

Files with missing lines	Patch %	Lines
src/implementations/truncation.jl	50.00%	10 Missing ⚠️

Files with missing lines	Coverage Δ
ext/MatrixAlgebraKitChainRulesCoreExt.jl	90.81% <ø> (ø)
src/algorithms.jl	80.70% <ø> (ø)
src/implementations/orthnull.jl	97.70% <ø> (ø)
src/interface/orthnull.jl	66.66% <ø> (ø)
src/pullbacks/polar.jl	100.00% <ø> (ø)
src/implementations/truncation.jl	80.88% <50.00%> (-1.81%)	⬇️

🚀 New features to boost your workflow:

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

[lkdvos]

Looks good to me!

Something I was wondering: do we want to extend this to an arbitrary number of truncation algorithms? Ie, something like

struct TruncationComposition{T<:Tuple{Vararg{TruncationStrategy}}}
    components::T
end
TruncationComposition(component::TruncationStrategy, components::TruncationStrategy) = TruncationComposition((component, components...))

It's not strictly necessary, since you could do (trunc1 & trunc2) & trunc3 as well, and I don't think the performance would matter all that much, just a design question I guess.

As a sidenote, it seems like the JuliaFormatter v2 changes are still not entirely without change. I'm fine with switching to the new version, but let's just make that a separate commit and rebase this on that?

[mtfishman]

Something I was wondering: do we want to extend this to an arbitrary number of truncation algorithms? Ie, something like

struct TruncationComposition{T<:Tuple{Vararg{TruncationStrategy}}}
    components::T
end
TruncationComposition(component::TruncationStrategy, components::TruncationStrategy) = TruncationComposition((component, components...))

It's not strictly necessary, since you could do (trunc1 & trunc2) & trunc3 as well, and I don't think the performance would matter all that much, just a design question I guess.

Mostly I did it the current way since as you say it is just as general but the implementation is a bit simpler. I can switch over to flattening, I guess the truncation strategies should commute with each other so order of operations doesn't matter and flattening is fine.

As a sidenote, it seems like the JuliaFormatter v2 changes are still not entirely without change. I'm fine with switching to the new version, but let's just make that a separate commit and rebase this on that?

Sure, will do.

[mtfishman]

Ok clearly I don't know how to use git, I don't know why the diff is showing all of the changes from previous PRs. But, I updated TruncationComposition based on your suggestion.

[mtfishman]

Maybe I'll just start a new PR from scratch...

[mtfishman]

Superseded by #18.