Stop the AI from sorting hands in place; unskip full-game tests

Under Emacs 28 and earlier, sort relinks a list destructively, so the
variable that held the old head can end up holding only a tail.  Four
computer-player choices sorted a list that shared structure with the
player's stored hand (cl-remove-if-not returns its argument when it
removes nothing), so Bridge, Briscola, Pitch and Euchre dropped cards
from hands mid-deal and then stopped on (= nil N).  Emacs 29's sort
reorders in place, which hid the bug there.

The previous commit skipped cgt-bridge-full and cgt-trick-ext-full
unless SVG is available, but both render text in batch; they failed
on Emacs 27 for this reason, not for want of SVG.

* card-games-bridge.el (card-games-bridge--ai-play):
* card-games-trick-ext.el (card-games-tx--ai)
(card-games-trick--ai-play) <pitch>: Sort a copy, never the hand.
* test/card-games-tests.el (cgt-bridge-full, cgt-trick-ext-full):
Drop the SVG skip.
(cgt--hands-snapshot, cgt-ai-play-leaves-hands-alone): New test; it
fails without the fix on Emacs 27 and on Emacs 29.
* NEWS: Note the fix.

Assisted-by: Mentat (for Corwin; claude-opus-5-5)

NEWS

@@ -3,6 +3,16 @@ card-games NEWS  -- user-visible changes
 
 * Version 1.0.92 (pretest3: MELPA review)
 
+** Bug fixes
+  - Bridge, Briscola, Pitch and Euchre no longer lose cards from a
+    hand part-way through a deal under Emacs 28 and earlier.  The
+    computer players sorted a list that shared structure with the
+    hand; ~sort~ before Emacs 29 relinks the list, which cut cards
+    out of the hand and later stopped play with a type error.  Emacs
+    29 and later were unaffected.  The full-game Bridge and trick
+    tests no longer skip without SVG, and a new test checks that a
+    computer player's choice leaves every hand unchanged.
+
 ** Packaging and internals
   - card-games-net requires subr-x.  It calls ~string-empty-p~, which
     lives in subr-x before Emacs 29, so a fresh Emacs 27 failed every