optimaltree hangs on edge case with all costs equal to 1

[lkdvos]

Fix for #206

@Jutho correct me if I'm wrong, but I think what is going on here is that the initial guess for the maxcost is wrong for these cases:
From what I can tell we use:

$$(\prod_i \chi_i) * \max_i(\chi_i)$$

Which evaluates to 1 in this case, while the actual maximal cost would be (n-1) * 1 with n the number of tensors.
I patched the specific case of all sizes equal to 1, but I am not sure if this now captures all possible problems. (I'm also not entirely sure where the initial formula came from?)

I could also replace this maxcost with the cost of simply multiplying in the order that is supplied, but I'm not sure what the implications are for the algorithm.

[Jutho]

I only remember that the initial cost was certainly trial and error, and not particularly clever. The cost of a simple left to right contraction would probably be much more sensible.

[lkdvos]

Test failures are unrelated, in principle this should be good to go.

[mtfishman]

Thanks for the quick fix @lkdvos! Can you register a new version?

[lkdvos]

Should be done relatively soon, but I want to include the precompilation PR as well