evil-ghostel: follow terminal cursor in emacs-state

[Cianidos]

Problem

evil-ghostel--around-redraw freezes point in every evil state except
insert, so entering emacs-state (the vanilla-Emacs escape hatch,
typically reached via C-z) leaves the cursor stuck wherever it was
when the state changed while the TUI keeps redrawing elsewhere. The
TUI's own rendered cursor (e.g. Claude Code's white prompt block)
moves, but Emacs's point doesn't, so cursor-sensitive operations and
on-screen feedback go out of sync.

Insert-state is the "follow cursor" state today — and emacs-state has
the same expectation from the user's perspective: it means "turn evil
off and behave like vanilla", which in a terminal context should mean
"follow the terminal cursor".

Reproduction

In any ghostel buffer with evil-ghostel-mode active, press C-z
to enter emacs-state. The Emacs cursor stays frozen while the
terminal keeps updating; moving the point manually is the only way
to "unstick" it. Switching back to insert-state re-enables the sync.

Fix

• Broaden the evil-ghostel--around-redraw freeze-exclusion from
  (not (eq evil-state 'insert)) to
  (not (memq evil-state '(insert emacs))) so redraws move point to
  the terminal cursor in both states.
• Add evil-ghostel--insert-state-entry to evil-emacs-state-entry-hook
  so switching into emacs-state syncs point immediately, matching the
  behaviour already in place for insert-state.

Includes a regression test
(evil-ghostel-test-redraw-moves-point-emacs-state) that enters
emacs-state, moves point away from the terminal cursor, forces a
redraw, and asserts point snaps back to the cursor column. Verified
to fail against upstream/main and pass on this branch.

[dakra]

Thanks