Prefer Gemini 3 over hardcoded Gemini 2.5 when available

[LyalinDotCom]

What happened?

Currently, several internal sub-agents, tools, and background services (such as the Codebase Investigator, CLI Help agent, summarizers, edit-corrector, and loop detection) rely on hardcoded model aliases that resolve specifically to gemini-2.5-flash or gemini-2.5-pro.

These internal tasks do not currently respect the user's broader "Preview Features" settings or model availability. Even if a user has opted into Gemini 3 Preview features and has access to the model, these background tasks remain locked to the older 2.5 variants. This creates a fragmented experience where the main chat might utilize the latest model (Gemini 3), but supporting logic and sub-agents degrade to the older generation.

What did you expect to happen?

The system should implement a centralized model resolution logic that automatically upgrades internal tasks to the "best available" model version based on the user's configuration and access rights.

Specifically:

1. Wherever gemini-2.5-flash is currently hardcoded or aliased for internal tasks, the system should check if the user has Gemini 3 Flash available (via Preview Features). If yes, it should use Gemini 3 Flash.
2. Wherever gemini-2.5-pro is currently hardcoded, the system should similarly check for and upgrade to Gemini 3 Pro if available.
3. This logic should be handled centrally so that toggling "Preview Features" instantly upgrades the intelligence of all sub-agents and tools without requiring individual configuration changes.

Client information

N/A

Login information

No response

Anything else we need to know?

• Exception for Flash-Lite: The alias gemini-2.5-flash-lite (used for lightweight tasks like simple summarization) should not be touched or upgraded at this time, as a "Flash-Lite" variant for Gemini 3 does not yet exist.
• Consistency: The goal is to align the internal "worker" models with the primary "driver" model selected by the user, ensuring that complex sub-tasks (like codebase investigation) benefit from the improved reasoning of the newer models.
• User Choice: If the user has Gemini 3 available but is selecting auto routing to 2.5, or manually selects a Gemini 3 model, we should use 2.5 models for sub agents and tools to respect their choice

[dhairyashiil]

I think this issue is already resolved by #18414 and #18773, so should we close this issue? please correct me if i am wrong @sehoon38 @LyalinDotCom