Expose tab and page selector controls to the agent

[cgasgarth]

Goal

Expose selector-style controls only where they are still required to unlock meaningful agent edits.

Current state

• The original iop/rgblevels/channel example is no longer a hard blocker.
• We already synthesize channel-specific and band-specific descriptors for rgblevels and colorzones, and the executor activates the needed selector internally before applying the edit.
• The generic catalog path still only emits descriptors for widgets it can map directly to set-float, set-choice, or set-bool.

Remaining gap

• Generic notebook/tab/page selectors are still not exposed as first-class capabilities.
• Modules that depend on selector changes without a bespoke synthetic binding are still effectively hidden from the agent.

Scope

• Decide when selector controls should be exposed directly versus handled through higher-level synthetic descriptors.
• Prioritize selectors that unlock otherwise unreachable edit settings, not cosmetic UI navigation.
• Keep capability and execution behavior deterministic enough for strict protocol validation.

Acceptance Criteria

• At least one additional selector-dependent module becomes agent-editable without requiring the model to drive raw UI navigation.
• Capability manifests and execution semantics clearly describe the selector-dependent behavior.
• The solution does not regress the existing rgblevels / colorzones synthetic descriptor coverage.