Drop the shipped example graph rather than ship one that no longer shows the pack

The example was written for an earlier node set. Yesterday it was rewritten inside a larger commit without
that being mentioned in the message, which is how it came to be a file nobody recognised. It shows nine of
the eleven nodes and neither of the two this release exists for, so it teaches the wrong shape of graph.

Removed with its generator and its test, since a test that validates a file that no longer exists is just a
red gate. Both are in the history if a new example wants them back. README says plainly that there is no
example right now and points at the LTX-2.5 recipe, which describes the same chain node by node.

`warm_demo.cube` goes with it. `nyc_skyline.png` stays, and the README now says why: it is the source image
the accuracy suite measures against in measure_ocio_parity.py, measure_histogram_compare.py and
gen_fixtures.py. Verified by hiding it and running the parity measure, which stopped with "could not read
nyc_skyline.png" - the published parity and histogram numbers are not reproducible without it.

Gate 32/32, one file fewer than before because the example test went with the example.

Author: SlavaSextron

README.md

@@ -783,14 +783,14 @@ through OCIO Write with the frame count intact.
 
 ## Example workflow
 
-`example_workflows/OCIO_Nodes.json` shows all eight nodes on one image. To open it:
-
-1. Copy `example_workflows/nyc_skyline.png` and `example_workflows/warm_demo.cube` into your **ComfyUI/input**
-   folder.
-2. Load `OCIO_Nodes.json` in ComfyUI and press **Queue**.
-
-You get the source image run through each of the six color nodes (with a preview each), plus an OCIO Read ->
-OCIO Write pair.
+There is no shipped example graph at the moment. The one that used to sit here was written for an earlier node
+set and would have to be rebuilt around the eleven nodes and the VAE path, which is worth doing properly
+rather than shipping a graph that opens and misleads. The LTX-2.5 recipe above describes that chain node by
+node, with a diagram.
+
+`example_workflows/nyc_skyline.png` stays, and not as a leftover: it is the source image the accuracy suite
+measures against, in `measure_ocio_parity.py`, `measure_histogram_compare.py` and `gen_fixtures.py`. Remove it
+and the published parity and histogram numbers stop being reproducible.
 
 ## Why this exists (and what's next)