fix: remove "--" from XML comments (forbidden by the XML spec)

The three notes added in this PR contained `--` inside `<!-- -->` blocks. The XML spec forbids that string in a comment. MuJoCo's TinyXML2 tolerates it, which is why the physics audit ran clean and the breakage was invisible from this side -- but every strict parser rejects the file, including `xml.etree.ElementTree`, which `mjcf_screwcap_to_usd.py` and the pack tests use. Rephrased to drop the token entirely rather than substituting an en dash, so the files stay pure ASCII. Verified: all 74 asset XMLs now parse with ElementTree; masses and the tennis-ball restitution are unchanged.

assets/box_task/feast_scented_candle/feast_scented_candle.xml

@@ -6,7 +6,7 @@
   action-replay through RoboVerse reproduces the recorded contact dynamics:
     - cylinder collision size 0.032 x 0.05 (radius x half-height), density 1219
 
-  2026-08-13 mass fix: density 120 compiled to 0.039 kg -- a scented candle in a glass jar
+  2026-08-13 mass fix: density 120 compiled to 0.039 kg, i.e. a scented candle in a glass jar
   weighing 39 g. Refit to 1219 kg/m^3 -> 0.40 kg over the 0.328 L collision volume.
   NOTE: 120 was copied from the Parrot scene so action-replay reproduced the recorded contact
   dynamics. This changes those dynamics; pack_box / place_candle_in_box demos recorded before

assets/local_meshy_cup_bottle/meshycup/mjcf/meshycup.xml

@@ -2,7 +2,7 @@
 <mujoco model="meshycup">
   <!-- Aligned with Parrot feast scene: density-only inertial, prop_collision (2026-05-15).
        2026-08-13 mass fix: meshycup_visual carried NO density="0", so MuJoCo weighed the render
-       mesh at its default 1000 kg/m^3 -- 0.639 kg of the compiled 0.677 kg, i.e. 94% of the mass
+       mesh at its default 1000 kg/m^3: 0.639 kg of the compiled 0.677 kg, i.e. 94% of the mass
        and the inertia came from a geom nothing can touch (its sibling meshybottle_visual was
        already tagged; this one was missed). Tagging it leaves only the shell, so the collision
        density goes 250 -> 1964 to land a 0.30 kg glazed ceramic cup over the 0.153 L shell. -->

assets/ycb/tennis_ball/tennis_ball.xml

@@ -8,12 +8,12 @@
   Density 358.7 kg/m^3 over the summed part volume 0.000162 m^3 -> mass 0.058 kg.
   Recentered AABB (m): size=[0.067, 0.067, 0.0665].
   2026-08-13 restitution: solref dampratio 1.0 (critically damped) made the one prop here that is
-  SUPPOSED to bounce perfectly inelastic -- measured e = 0.000 against the ITF spec e = 0.73
+  SUPPOSED to bounce perfectly inelastic; measured e = 0.000 against the ITF spec e = 0.73
   (1.35 m rebound from 2.54 m). Authored 0.006 s / 0.30 gives a consistent e = 0.39 (0.40, 0.46
   bounce-to-bounce). 0.73 is NOT reachable in this contact model at dt=0.002: below dampratio
-  0.28 the ball GAINS energy between bounces (e = 1.32, 1.63 measured) -- numerical, not physical.
+  0.28 the ball GAINS energy between bounces (e = 1.32, 1.63 measured): numerical, not physical.
   NOTE: HAND_TRACK_SIM_PARAMS overwrites every geom solref, so this needs a backend opt-out to
-  take effect in a shipped run; audited with --profile authored.
+  take effect in a shipped run; audited with the "authored" contact profile.
   License: YCB CC BY 4.0.
 -->
 <mujoco model="ycb_tennis_ball">