unfavorite kinova_sim Velocity Force Controller Zero, and zero fts

Replace `GetMasks2DFromTextQuery` (CLIPSeg) with `GetMasks2DFromExemplar`
(SAM3 multimodal segmenter) in the `ML Move Boxes to Loading Zone`
Objective and its inner Subtree. SAM3 also accepts a text prompt but
returns SUCCESS with `mask_count=0` on empty detections, letting the
Subtree skip the downstream pick pipeline cleanly with `_skipIf`.

Restructure the outer loop from the `RetryUntilSuccessful + ForceFailure`
infinite-loop pattern to
`Fallback { RepeatUnlessFailureEachTick { ... ScriptCondition(!drained) }, ScriptCondition(drained) }`
so the loop exits with Objective SUCCESS when the scene is drained but
surfaces real pick errors as Objective FAILURE.

Fixes #19112.

Co-Authored-By: Claude Opus 4.7 (1M context) <[email protected]>

…Boxes

Updates the View Boxes text prompt from "brown cube" to
"a single small orange cube box" to better describe the sim assets,
then guards the SAM3 output with FilterMasks2DByArea so the
downstream pick pipeline only sees plausibly-sized box masks:
the cart frame measured ~26,285 px in calibration and was the
primary false positive driving wasted pick attempts; thin
edge-fragment masks measured ~778 px. Real boxes in this scene
measure 3,949-20,643 px, so thresholds are min=2,000 / max=23,000.

GetSizeOfVector refreshes {mask_count} after the filter so the
existing `_skipIf="mask_count == 0"` gate reflects the filtered
count rather than SAM3's raw count.

Co-Authored-By: Claude Opus 4.7 (1M context) <[email protected]>

…-filter

Fix: Filter SAM3 mask false positives by area in hangar_sim

…sam3

Fix: Use SAM3 instead of CLIPSeg in ML Move Boxes Objective

bump: phoebe_ws to get fixes

add a wait to give TF time to be available

Pulls in PickNikRobotics/picknik_accessories#45 — makes the lab_sim
burner pushable under realistic gripper force.

…ries-burner-v9.3

Fix: Bump picknik_accessories for lab_sim burner movable fix