Proper Internal Error Reporting

[quantumish]

Right now, internal server errors are propagated until they are blindly .unwrap()ed in the top level request function, which is annoying for a number of reasons:

• End users get a "connection reset" error when they shouldn't
• Users can't properly handle internal errors and have to resort to checking stdout (not even a log file!)
• It's unclear what's causing the error for a developer: as even the logged error in stdout doesn't have a backtrace (and RUST_BACKTRACE=1 backtraces are too big)

The main problem is that either cassandra_cpp::Result or generic error reporting libraries like anyhow or error_stack aren't thread-safe, making it impossible to use in the server.