feat: remove by ID only skeleton source

[seankmartin]

This skeleton source for catmaid was when we pulled full skeletons and spatial chunks
through different sources. Since we unified them it doesn't have much purpose. You can
just set the hidden skeletons to be transparent and then they won't fetch and you can
pull by ID.

[seankmartin]

This will change existing states if they listed a change to the sources but I don't expect that to happen. We may want to try merge this into neuroglass before going to prod though, or shortly after. If we agree on removing this source

[afonsobspinto]

I understand. I don't think it's a problem either because it was demo and I agree as well that if we deploy to prod without this change then it becomes a problem

[afonsobspinto]

I agree with the change. While having codex review the code it mentioned something I though we already fixed but maybe we didn't:

This is unchanged code, but it matters when removing the alternative source. With visible opacity 1 and hidden opacity 0:
- The backend stops requesting spatial chunks.
- areVisibleChunksReady() still requires those chunks to reach GPU memory.
On a fresh view—or after moving into an uncached region—the layer can therefore remain unready even after the selected skeleton loads. [Readiness check](https://github.com/MetaCell/neuroglancer/blob/26c42415ed90efb62bad45f640c13a7b0aef148f/src/skeleton/frontend.ts#L3018)
This affects screenshot capture: normal capture waits for viewer.isReady(), although the UI has a timeout-based forced-capture fallback. [Screenshot readiness gate](https://github.com/MetaCell/neuroglancer/blob/26c42415ed90efb62bad45f640c13a7b0aef148f/src/python_integration/screenshots.ts#L179)
Suggested fix: make readiness follow the rendered passes—skip spatial-chunk requirements when hidden opacity is zero, and account for selected skeletons’ full-data loading.