Some queries weren't using the full power of SQL and doing some of
the selection in the code, fix that (by_hash and host mapping).

This will help keep popular tags near the top, while expiring less
used tags

My testing machine has 10 cpus, and so didn't trigger the failure
where not all the work was submitted..  We need to pop the completed
work items, and keep doing the for loop while we have futures to
process...  this submits and processes all work..

as we might have a lot of work to submit, BUT it might fail early,
don't submit too much work early on, just for it to fail, so we
limit how much work we submit..

This will be used to allow parallel processing of torrent pieces..

each piece of the torrent can be processed in parallel, and this
class will make sure that when processing the hash of a file in
the torrent, it will be hashed in the correct order...

setdefault works best for simple data types, calling the factory
function every time doesn't make sense...

hide coverage in test code

Hard code stat results so that the tests pass on a different checkout
w/ different mtimes..

Looks like this is the only way:
https://github.com/sqlalchemy/alembic/discussions/1086

add tests for migration, good example for future needs..

We were only using the ed448 code, and cryptography requires rust
to compile, and contains a LOT of other code, while edgold is much
smaller, and doesn't require a rust compiler, and will work on more
systems..  Keep around an old identity generated from cryptography
to make sure the two libraries are compatible.