feat: converter keeps layers of screens and camera, perspective, matrix calls; That New Teacher runs its own text-tag libraries

onlayer on show/hide/call screen and `camera <layer>` reach the player, `perspective`
is an ATL property now, RotateMatrix/OffsetMatrix/ScaleMatrix, Text.set_text and
renpy.fetch are accepted. The text-effect modules of That New Teacher are no longer
skipped (only the shader-based wave tag is).

Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>

backend/app/converter/atl.py

@@ -43,7 +43,7 @@ LEADING_ZERO = re.compile(r"(?<![\w.])0+(\d)(?![\d.]*[\w.])")
 SPLIT_SIGN = re.compile(r"^([a-z_]+)(-[\d.]+)$")
 IGNORED = {"subpixel", "animation", "around", "alignaround", "rotate_pad"}  # no visible effect in the browser
 # Purely graphical GPU hints; dropping them changes how a picture is drawn, never what happens
-COSMETIC = ("shader", "mesh", "gl_", "u_", "perspective", "point_to", "delay", "events", "nearest", "fps", "show_cancels_hide")
+COSMETIC = ("shader", "mesh", "gl_", "u_", "point_to", "delay", "events", "nearest", "fps", "show_cancels_hide")
 NAMED_POSITIONS = {
     "left": {"xpos": 0.0, "xanchor": 0.0}, "right": {"xpos": 1.0, "xanchor": 1.0},
     "center": {"xpos": 0.5, "xanchor": 0.5}, "truecenter": {"xpos": 0.5, "xanchor": 0.5, "ypos": 0.5, "yanchor": 0.5},