Fix FormNetwork constructor bug

[JoeyT1994]

This PR fixes some bugs in the constructor of the identity tensors in the middle of a QuadarticFormNetwork or BiLinearFormNetwork which emerge when quantum numbers are enabled.

Specifically, denseblocks is cast on the identities on each site to ensure they can be multiplied together and also the order of the indices is corrected so that the appropriate identity matrix is constructed when enable_auto_fermion is active.

@emstoudenmire @anmello hopefully this is sufficient to fix some of those fermionic issues that were arising.

[emstoudenmire]

Thanks, Joey. Looks good & this reordering of indices fixed the fermion bug I had.

[mtfishman]

Actually I merged prematurely. I think this should be:

Suggested change

	return denseblocks(delta(last(i_pair), dag(first(i_pair))))
	return denseblocks(delta(Bool, last(i_pair), dag(first(i_pair))))

Otherwise, delta will make a tensor with element type Float64, which will then promote other tensors to double precision, which is an issue on GPU where it is generally better to use single precision. @JoeyT1994 can you make a new PR adding back Bool, or did you see some issue with that?

[JoeyT1994]

@mtfishman This introduces bugs to me that I don't understand. For instance

    g = named_grid((4,))
    s1, s2 = siteinds("S=1/2", g), siteinds("S=1/2", g)

    sind1, sind2 = only(s1[(1,)]), only(s2[(1,)])
    ITensorNetworks.itensor_identity_map([sind1=>prime(sind1), sind2=>prime(sind2)])

returns

Whilst

    #More complex constructor 
    g = named_grid((4,))
    s1, s2 = siteinds("S=1/2", g; conserve_qns =true), siteinds("S=1/2", g; conserve_qns =true)

    sind1, sind2 = only(s1[(1,)]), only(s2[(1,)])
    ITensorNetworks.itensor_identity_map([sind1=>prime(sind1), sind2=>prime(sind2)])
    

returns

[JoeyT1994]

If there's a way to avoid these bugs and avoid making it Float64 type I can make that fix in an new PR

[mtfishman]

I see, interesting. An alternative design could be:

function itensor_identity_map(elt::Type, i_pairs::Vector)
  return prod(i_pairs; init=ITensor(one(elt))) do i_pair
    return denseblocks(delta(elt, last(i_pair), dag(first(i_pair))))
  end
end

itensor_identity_map(i_pairs::Vector) = itensor_identity_map(Float64, i_pairs)

Then we can set the element type as needed.

[mtfishman]

(Though it would be nice to fix that bug anyway.)