feat(settings): support preferred text size / Dynamic Type

[stew-codex]

Problem

agent-device settings does not expose the preferred text-size / Dynamic Type category. This makes accessibility and layout QA less reproducible: an agent can observe unexpectedly large text, but it cannot inspect or restore the simulator's content-size category through the agent-device command surface.

Observed with agent-device 0.21.0 on an iOS simulator. The current workaround is to leave agent-device and call CoreSimulator directly:

xcrun simctl ui <udid> content_size
xcrun simctl ui <udid> content_size large

Driving the Settings app is possible, but it is slower and depends on localized labels and slider coordinates.

Proposed command

Add an iOS-simulator settings helper, for example:

agent-device settings text-size
agent-device settings text-size large
agent-device settings text-size accessibility-extra-large

The no-argument form should return the normalized current category. The setter should accept the categories supported by simctl ui ... content_size:

• extra-small
• small
• medium
• large
• extra-large
• extra-extra-large
• extra-extra-extra-large
• accessibility-medium
• accessibility-large
• accessibility-extra-large
• accessibility-extra-extra-large
• accessibility-extra-extra-extra-large

Exact naming can follow the existing settings-command conventions.

Scope

• Implement the first slice for iOS Simulator by delegating to simctl ui <device> content_size.
• Respect the active session or explicit device selection (--device / --udid).
• Return UNSUPPORTED_OPERATION with a useful message on targets where agent-device cannot set the category.
• Android font-scale parity can be handled separately because its values and behavior do not map exactly to Apple's categories.

Acceptance criteria

• The current preferred content-size category can be read through CLI and structured API/MCP responses.
• A supported category can be set on the selected iOS simulator.
• The result reports the applied normalized category rather than only command success.
• Invalid category values fail before mutating simulator state and list the accepted values.
• Help text and command documentation include the new setting.
• Tests cover parsing, device selection, success, invalid input, and unsupported-target behavior.

[KashmirAwana]

It sounds like the current agent-device command doesn't support adjusting text size for accessibility testing on iOS simulators. I'd check how the proposed commands would integrate with existing settings.