Fix combining interfaces

[mtfishman]

This fixes combining interfaces with more than two objects.

@lkdvos at some point I think you brought up making the design of this system closer to the design of BroadcastStyle (i.e. the interface for defining and combining interfaces), maybe I'll try that out here as well.

[codecov]

Codecov Report

Attention: Patch coverage is 45.45455% with 6 lines in your changes missing coverage. Please review.

Project coverage is 72.44%. Comparing base (8371a5e) to head (3ce69a4).
Report is 1 commits behind head on main.

Files with missing lines	Patch %	Lines
src/abstractinterface.jl	40.00%	6 Missing ⚠️

Additional details and impacted files

@@            Coverage Diff             @@
##             main      #29      +/-   ##
==========================================
- Coverage   72.89%   72.44%   -0.46%     
==========================================
  Files           9        9              
  Lines         321      323       +2     
==========================================
  Hits          234      234              
- Misses         87       89       +2     

Flag	Coverage Δ
docs	37.18% <9.09%> (-0.55%)	⬇️

Flags with carried forward coverage won't be shown. Click here to find out more.

☔ View full report in Codecov by Sentry.
📢 Have feedback on the report? Share it here.

[lkdvos]

Yeah, it would be nice to have:

• symmetric promotion rules where you only need to define one
• a clear entry point, what do we need to overload?
• a way to signal that no promotion could be found

[mtfishman]

Gotcha, I'll try it out. Partially I was pinging you to make sure this is orthogonal to what you're working on in #28, which I think it is.

[mtfishman]

@lkdvos I decided to make this PR more incremental, since I think more ambitious plans would require some breaking changes and also lead to conflicts with #28, so we are better off merging #28 first and then looking into more ambitious changes to this part of the code after that. But, this PR is still helpful since it fixes a bug I was hitting when broadcasting NamedDimsArrays wrapping BlockSparseArrays (see ITensor/NamedDimsArrays.jl#47), and also makes the code a bit clearer.

Basically, the plan in a future PR would be to bring this more in line with the design of Base.Broadcast.BroadcastStyle, which would mean that:

1. types would overload DerivableInterfaces.AbstractInterface(::Type) to define their interface as opposed to the current DerivableInterfaces.interface(::Type) syntax,
2. interface combining rules would be defined through overloading DerivableInterfaces.AbstractInterface(::AbstractInterface, ::AbstractInterface) as opposed to the current DerivableInterfaces.combine_interface_rule(::AbstractInterface, ::AbstractInterface) syntax, and
3. combine_interfaces(x...) would be used to get the combined interface of a number of objects/arguments, as opposed to the current interface(x...) syntax.

Base.Broadcast seems to have made some reasonable choices so it seems like we may as well just follow what they did for this and other similar trait systems. Curious what you think.

[lkdvos]

I think when I was looking at it I had a similar feeling, that the Base approach was definitely reasonable. What you are describing sounds great, and since it aligns with Base that also removes the mental bandwidth of having to remember two different systems so I'm definitely in favor of this.

Changes here also look good, it's convenient that it's not breaking, and should be easy to rebase the other PR on top of.