Improve handling of RGB dataset dimensions

[axelboc]

Fix #1838

This PR fixes the issue above by now sending a complete selection string for RGB stacks with all providers, therefore moving past the fetching error in HSDS.

This is done by making a more general improvement to how dimensions of RGB datasets are handled: when the RGB vis is selected, the last dimension (the one with the channels) is now visible in the dimension mapper but locked so it cannot be mapped to an axis:

Screencast.from.2025-08-14.09-29-05.webm

---

I've done this by introducing the concept of "locked dimensions" to the dimension mapper. When a dimension is locked, it is marked as null in the dim mapping state.

[axelboc]

The screenshot and recording were actually taken with h5grove because I have another issue with HSDS: the RGB images are now properly fetched with full selection strings, but they're all black because I'm only getting zeros... I'm not sure if it's specific to the file I'm testing with (cf. #1838) or something in HSDS. We're not on the latest version, so I'm hoping @Flydroid won't get the issue. Otherwise, I'll open a dedicated issue.

Note that the mandrill image still works fine with HSDS and that you can now see the last dimension:

[axelboc]

/approve

[loichuder]

the RGB images are now properly fetched with full selection strings, but they're all black because I'm only getting zeros...

Probably due to some mismatch in the expected format. If you expect RGB integers (0-255), but get RGB floats (0-1), the numbers will all be zeros.