Improve parallelisation

[RMeli]

PR #113 introduced parallelisation of rmsdwrapper, with the goal of providing a very useful timeout functionality (to avoid hangs in pathological cases where graph isomorphism takes a very long time).

However, the parallelisation over each call is inefficient since the expensive graph isomorphism is computed repeatedly. https://github.com/RMeli/spyrmsd/pull/113#issuecomment-2181394583 describes two potential avenues for more efficient and sensible parallelisation.

Computing the graph isomorphism serially (with timeout) and parallelising over RMSD calculations would also avoid possible clashes with internal parallelisation of the graph isomorphism algorithms.