docs(model): disambiguate vision_context_parallel from dist-train knob

Bridge already exposes dist_train.vision_context_parallel_size, an
unrelated Qwen3-VL knob that gives the vision tower its own ranks and
sets the encoder's own context_parallel_size. Add a comparison table so
the two are not mistaken for alternatives, and record that the encoder
config intentionally keeps context_parallel_size=1 under this feature
because the sharding happens outside the encoder, over its inputs.

Drops the README link into the skill tree per review feedback.

Signed-off-by: Huy Vu <huvu@nvidia.com>

Author: huvunvidia

examples/models/nemotron/nemotron_3_omni/README.md

@@ -292,9 +292,6 @@ Multi-image data whose per-sample image count varies (Mantis at CP=4) therefore
 balances less evenly than uniform data, which is why its memory saving is the
 smallest of the three.
 
-For code anchors, pitfalls, and the A/B verification procedure, see
-[the vision context parallel skill](../../../../skills/nemo-mbridge-perf-vision-context-parallel/SKILL.md).
-
 ### Image-Text — CORD-V2
 
 [CORD-V2](https://huggingface.co/datasets/naver-clova-ix/cord-v2) is a

skills/nemo-mbridge-perf-vision-context-parallel/SKILL.md

@@ -22,6 +22,22 @@ borrowing the CP group** — a rank encodes whole images, never partial ones.
 Shipped for Nemotron Omni. The MCore split/gather helpers are model-agnostic, so
 the pattern generalizes to any Bridge VLM with a dynamic-resolution encoder.
 
+## Do Not Confuse With `dist_train.vision_context_parallel_size`
+
+Bridge already has a similarly named and completely different knob:
+`vision_context_parallel_size` on the dist-train path
+(`src/megatron/bridge/models/qwen_vl/qwen3_vl_provider.py`, consumed in
+`src/megatron/bridge/training/initialize.py`).
+
+| | `model.vision_context_parallel` (bool) | `dist_train.vision_context_parallel_size` (int) |
+|---|---|---|
+| Ranks | Reuses the language model's CP ranks | Gives the vision tower its **own** ranks |
+| Splits | Whole images across those ranks | The vision encoder's own sequence dimension |
+| Encoder `context_parallel_size` | Stays `1` | Set to this value |
+| Models | Nemotron Omni | Qwen3-VL dist-train |
+
+They are not alternatives and do not compose in any tested configuration.
+
 ## Enablement
 
 ```python
@@ -95,7 +111,10 @@ Upstream helpers (do not modify — changes go through the MCore repo):
    images so no rank runs an empty encoder, and returns `num_padded_ranks` so the
    gather can drop them again.
 3. **Encode.** Each rank runs the full vision tower on its own images only.
-   Activations, not just compute, are what drop by `1/CP`.
+   Activations, not just compute, are what drop by `1/CP`. The vision encoder's
+   own `context_parallel_size` stays hardcoded to `1` in
+   `_build_vision_config` — that is correct and unchanged by this feature, since
+   the sharding happens *outside* the encoder, over its inputs.
 4. **Even-grid padding.** `_pad_patch_grid_to_even` zero-pads an odd patch grid
    before pixel shuffle. Only the injected `1x1` placeholders hit this path;
    real frames are already even in both extents.