io

Added those in. I'm still converting to String with getCanonicalName in the getLogger(Class ...) method because, e.g.
JLogger jlogger1 = JLogger.getLogger(Jalview.class);
JLogger jlogger2 = JLogger.getLogger("jalview.bin.Jalview");
should really return an existing logger in the registry if it exists, and the String name is the registry key.

I was thinking that JLogger did this comparison anyway, but I've followed it through and it didn't. So I've added

if (loglevel.compareTo(this.level) < 0) { return false; }

to JLogger.println so that a comparison is made between the (enums) of JLogger.level and the message LogLevel, and if it's not going to be logged then it doesn't call the actual logger at all (in this case Log4j, so the log4j object is never even touched [even with logger.isDebugEnabled() because the answer should be no]).
This now applies to all log calls.
The check whether Cache.log != null is equivalent to Cache.log.loggerExists() which already happens within JLoggerLog4j.
There is admittedly the wrapper of the JLogger getting brought to forefront of memory, but that's presumably smaller than the old Cache.log which was a Log4j.Logger object.

AWESOME JOB !!!! one tiny niggle I had was that Cache.debug/Cache.error doesn't read nicely, so I've abstracted the logging stuff even farther to a new jalview.bin.Console class: see https://source.jalview.org/crucible/changelog/~br=update%402fJAL-3949_jalview_bin_Console_to_cure_sanity_even_more/jalview/

the reason for conditionals is that Java won't perform the string append operation if logLevel>debug. It saves a minor amount of processing time which is important for code that is executed lots.

You could have .getLogger(Class ..) rather than having to call getCanonicalName ?