IndexSet set operation interface improvements

[mtfishman]

Many IndexSet set operations have first semantics, i.e. they return the first Index found but don't error if more than one are found.

It would be nice to have another set of functions that have only semantics, which error if there is not exactly one Index that satisfies the specified requirement.

For example:

i = Index(2; tags="i")
is = IndexSet(i, i')
firstind(is; tags="i")  # returns i
onlyind(is; tags="i")  # would error

is2 = IndexSet(i, i', i'')
commonind(is, is2)  # returns i
onlycommonind(is, is2)  # would error

It would also be nice to make some new aliases:

commonind(_) = firstcommonind(_)
ind(_) = firstind(_)
[...]

so that each set function has versions first*ind(_), only*ind(_), and an alias *ind(_) = first*ind(_) for convenience.

This would be useful in places where you know that there should be one and only one Index that is shared, for example in MPO functions where there should only be one site Index shared between two MPO tensors.

We could also consider giving the *ind(_) functions a keyword argument, like *ind(_; only = true), which would be another syntax for only*ind(_).

[mtfishman]

Adding onto this with a related point, it might be nice to add an alias filterinds([...]) for the function inds(::ITensor; kwargs...) (such as inds(A; plev = 0, tags = "i")) which filters a subset of indices based on the criteria that are provided. That is just to make the name more descriptive than the current name inds.

For all of the IndexSet set operations that involve multiple indices, like commoninds(...), uniqueinds(...), etc., it would also be nice to have versions *inds(::Order{N}, ...) where Order{N} would be a value type we would define that represents the compile-time order of a tensor. This would help in many places where you know what the order is supposed to be and want type stability (or a check that the order is correct).