Simplify similarmatrix, remove optimize!

[pablosanjose]

Closes #105

This PR is an alternative to #105

PR #105 has some serious problems. Internalizing the Bloch matrix into Hamiltonian leads to less flexibility, since its type often depends on the context (e.g. the method used to diagonalize a Hamiltonian). Needing to mutate/regenerate a Hamiltonian just to change its Bloch matrix type is inelegant and inefficient. Also, a chain of transformations of a Hamiltonian almost always led to a blochmatrix needing to be allocated at each step. The decision to take the Julian bloch!(matrix, h, ...) approach was ultimately good. The only problem of needing an external preallocated Bloch matrix is that the similarmatrix used to construct the matrix was sometimes not smart enough.

This PR takes the good bits of #105 without internalizing blochmatrix. In particular we can now do similarmatrix(h, flatten) to keep the matrix type of h but with a scalar eltype (it is still inferable since flatten is a function with a unique type). Since flatten is an exported function with the same meaning of flattening eltypes, this seems appropriate, without needing a new exported function. We also get rid of optimize!, and simply force the output of similarmatrix(h,...) to already be optimized for h, thus avoiding sparse element splicing when calling bloch!/bloch.

We also sneak in a change that was part of #105: changing the type kwarg of hamiltonian to orbtype, which is more descriptive. The equivalent for lattice is still type.

[codecov-commenter]

Welcome to Codecov 🎉

Once you merge this PR into your default branch, you're all set! Codecov will compare coverage reports and display results in all future pull requests.

Thanks for integrating Codecov - We've got you covered ☂️