Redesign argument storing

* Use state objects instead of tons of locals in the main
  function.
* Use actions for storing arguments into result table.
  Actions are now called at the end of each invocation,
  with result table, target index, arguments and overwrite flag as
  arguments.
* Remove command actions.
* Improve error messages, refer to options by the last used alias
  instead of the main name.

TODO:

* Improve error messages further ("argument 'foo' is required"
  -> "missing argument 'foo'", etc.).
* Add actions for positional arguments.
* Add actions for commands (should be called with final results
  after parsing is over, in "innermost first" order).
* Allow referring to built-in actions by strings a-la Python
  (e.g. action = "store_false").
* Allow setting initial value to be stored at target index
  for each option (perhaps use default value for that).
* Add more tests, particularly for actions.

spec/integrity_spec.lua

@@ -20,7 +20,7 @@ describe("tests related to CLI behaviour #unsafe", function()
          assert.equal([[
 Usage: ]]..script..[[ [-v] [-h] <input> [<command>] ...
 
-Error: too few arguments
+Error: argument 'input' is required
 ]], get_output(""))
       end)
 
@@ -62,7 +62,7 @@ Did you mean 'install'?
          assert.equal([[
 Usage: ]]..script..[[ install [-f <from>] [-h] <rock> [<version>]
 
-Error: too few arguments
+Error: argument 'rock' is required
 ]], get_output("foo install"))
       end)