Move base level code to an activate function

Liquidprompt has always had a good chunk of its startup logic at the
root level, not in functions. This code is run only once, when it is
sourced. Part of the problem with this is that if we want to test
Liquidprompt, we need to source it, but we don't want to load all the
actual prompt setup stuff, only get the functions.

This breaks out the activation stuff into a lp_activate function, so
that Liquidprompt can be sourced without running any of that. Now if the
'--no-activate' flag is given, it isn't run.

The other problem is that a Liquidprompt configuration change requires a
full source of Liquidprompt again to update to that configuration. With
a correctly built activate function, the config can be reloaded and
Liquidprompt reconfigured without needing to source again.

To make enabling features work with a re-run of lp_activate(), we need
to remove the checks that only define functions if their respective
feature is enabled. This does not slow down startup time, but it does
slightly pollute the listing of function definitions. Since all of these
functions are prefixed with '_lp', users shouldn't see them unless they
are looking for them.

LP_TIME is broken in this commit: the way it is written now, the
"correct" _lp_time() function is set at the root level, after config
loading. But now config loading happens last, so no _lp_time() is set
(other than the empty one). Anyway, it needs a bunch more work to fix
this, and I don't want a huge commit.