Parallelize component loading/parsing

We might as well load and parse the individual components in parallel. Most of the code does not use Federations, so it is thread safe already. When #111 is complete we can revisit and parallelize even more code.

I suspect we can gain some performance improvements here, but we should benchmark the multithreaded implementation to ensure that it is indeed faster (the minor overhead may outweigh the benefits in some cases).

If we can parse/load components very quickly we may even be able to move to a stateless Reveaal where we maintain no state between queries. This allows us to do other parallelization more easily as there is no shared state between queries.

Start by locating the component loading and parsing implementation and establish a benchmark for the current behavior. Compare it with a multithreaded implementation, including cases involving Federations, and consider the stateless approach described in the issue. Done means the benchmark shows whether parallelization improves performance without breaking component handling.

Written by the indexing model from the issue text.