Fixes for recently-introduced readline and $&time bugs

[jpco]

Fixes #151 (introduced with #128) and #152 (introduced with #143).

getrusage() has been pulled out of proc.c entirely and is now managed within $&time. I think this structure is a lot better in terms of separation-of-concerns. It's also nicely simple to implement with how getrusage() works, which is something that seemed very weird to me at first. So that's nice!

It occurs to me with this change that if we make use of getrusage(RUSAGE_SELF) as well, we could actually $&time commands without forking, which seems like an interesting possibility. If you have a slow-running script, you could just wrap whole sections inside a time {} without any disruption to the script's function... Even better, it looks like the standard times() also supports getting times from the calling process, so we could keep things consistent. (Edit: This is now #154)

We don't do anything so clever with rl_reset_screen_size(), even though it's easy to imagine something that waits for a SIGWINCH and only calls resets the screen size then. However, that requires a novel kind of special signal handling for SIGWINCH, and SIGWINCH isn't even technically POSIX.1-2001 standard, so we'd have to wrap a fair amount inside of #ifdef SIGWINCH, and this is all adding a fair amount of complexity for a very negligible performance/efficiency gain.