Intersphinx could fetch remote inventories concurrently

[Zac-HD]

Hypothesis has optional integrations with many other libraries, and our intersphinx config currently links to six other sets of documentation. This works beautifully, and we all really appreciate Sphinx! ❤️

However, when the cache is cold or doing a clean build, fetching all of these inventories can take considerably longer than the rest of the build. I propose to use a concurrent.futures thread pool to run the inner loop in load_mappings() when there is more than one inventory that needs to be fetched.

This could give slightly better or slightly worse performance for all-local inventories, but will be much faster for remote inventories. I am happy to open a PR with this change.

[tk0miya]

+1; reasonable. PRs are always welcome!