performance improvements

[anttikantee]

While fs-utils runs in well under 100ms on a modern PC, execution time for a simple ls can be 300ms on RPi. This is too slow for comfortable use. While this issue is not high-priority, I'm writing down some ideas which need to be evaluated.

• if file system type is not specified, try to mount file systems in some sensible order of popularity
• don't link in any file system modules at compile-time. instead, dlopen() them at runtime. this avoids unconditionally running a bunch of unnecessary module constructors with every invocation
• extract probe routines from the drivers into standalone routines at least for the most common file systems. this avoids having to load a module before being able to determine if that driver can handle a given image
• profile the various fsu utils -- a process which only does rump_init() with -lrumpvfs -lrump runs in ~70ms on the RPi. we're spending 230ms somewhere else.