This causes issues with the player: prefix used in any() and all()
constructs, as player: will now be parsed as a word instead of the
proper token. For now, : is disallowed in words again until there is a
better solution.

Away with stringly typed stuff, we now have a proper way to save the
profession of a player without relying on a string. Theoretically, that
is better for memory consumption, as we now save only the identifier and
use fmt::Display and static strings, but that was not the main reason
for this change.

The main reason is that now we can programatically work with the
profession and elite spec, so that we can (for example) implement a
filter to filter players based on their class.

The word "class" has been chosen because it is a common synonym for the
profession/elite, and because this is neither a profession nor the elite
- it's a combination of both.

Rust doesn't necessarily need this, but it's good formatting to include
it anyway.

With the file name heuristic for -before and -after in place, we might
want a way for the user to disable it. For now, we simply do this by
providing a new set of predicates without the filter.

In the future, we might have a --disable-heuristics switch to disable
the heuristics, in case we ever add more.

This avoids unwrapping (and therefore panicing) when the path doens't
have a parent. It also avoids the explicit Into::into() calls.

This persists the REPL history across program restarts.

The code should probably be cleaned up a bit more, the error handling in
this one is a bit all over the place. This is because we don't want to
make it a hard error in case the history cannot be saved.

With evtclib being on crates.io now (and generally, being available
publicly through git), there is no longer a need to have a copy of the
evtclib repository in this repository. The main reason was that evtclib
was private when I started it, so the easiest way to include it was
through a git submodule. That reason is no longer valid.

If we really *need* to use the git version, it is also better to just
point Cargo to the repository and let it deal with keeping the
repository up-to-date, rather than using git submodules.

This commit also updates Cargo.lock, so there is a bit of noise from
also adding ctrlc as a dependency.

This is not yet perfect, as it seems to still execute all queued threads
just to immediately exit them, so maybe we should try and see if we can
"clear" the rayon queue. But it's a good start, and the ctrlc crate
seems to work well for this job.