Fix some bugs related to repeating company-mode completion commands

[bcc32]

Repeating company-complete using evil-repeat doesn't work
correctly without some workarounds.

Prior to this PR, I identified two bugs that occur, see below for
reproduction. This change is intended to fix both issues.

When there is a common prefix to complete to

In the *scratch* buffer in lisp-interaction-mode:

(company-mode)
(evil-mode)

(keymap-set company-mode-map "TAB" #'company-complete)

;; Start at the end of the first "req" and hit TAB, which completes to
;; "require".  Hit "j ." and see that repeat doesn't work correctly.
(req)
(req)
(req)

;; Then, evaluate this and see that the above completions work
;; correctly.
(evil-declare-change-repeat #'company-complete)

When there is no common prefix

In the *scratch* buffer in lisp-interaction-mode:

(company-mode)
(evil-mode)

(keymap-set company-mode-map "TAB" #'company-complete)

;; Start at the end of the first "re" and hit TAB, then pick some completion.
;; Observe this error in the *Messages* buffer:

;; Error in post-command-hook (evil-repeat-post-hook): (wrong-type-argument number-or-marker-p nil)

(re)
(re)
(re)

[bcc32]

For context, the post-command-hook issue happens because company-complete uses call-interactively and that causes this-command to be different betwen the pre- and post-hooks. See company-mode/company-mode#383