Adjoint transfer tests are incorrect

[jwallwork23]

If different meshes are used instead of the trivial ones currently used in TestTransfer then we get test failures for the adjoint transfer operators. The tests assert that applying the adjoint transfer to a source should give the same result as a forward transfer, but this is nonsense.

[jwallwork23]

Two of the tests are actually fine (adjoint transfers between matching spaces) and are re-enabled in #187.

However, the tests for different spaces on the same mesh aren't and need more work. We should write down what the transfer matrices look like for very small, simple examples and use those to work out what the adjoints should be.