Auto model selection should offer quality, balanced, and economy routing profiles

## Feature request: configurable Auto routing profiles

Please add user-selectable routing preferences for **Auto** in VS Code Copilot Chat and the Agents experience:

- **Quality** — favor the model expected to provide the strongest result for the task, accepting higher cost or latency when needed.
- **Balanced** — choose the best overall quality, cost, and responsiveness trade-off.
- **Economy** — favor lower cost and faster responses when the expected task quality remains sufficient.

The current Auto experience can appear to route repeated coding tasks through a narrow subset of models, while other models available to the user may offer a better quality/cost trade-off for that task. Users need a supported way to state which trade-off matters to them without manually choosing a model for every prompt.

## Expected behavior

For each profile, Auto should select from the current models eligible for the user's plan and organization policies, subject to real-time availability and reliability. Its decision should consider:

- the request type and complexity, such as coding, debugging, editing, research, or simple lookup;
- task fit and model capabilities, including tool and context compatibility;
- current model lifecycle and availability, as models are introduced, changed, or retired;
- the relevant quality, cost, latency, and reliability trade-off for the selected profile.

This is not a request for a fixed model ranking or a hardcoded vendor preference. Auto should keep using server-side routing and fallbacks, but it should use the current eligible catalog and honor the user's chosen routing objective.

## Subagents and transparency

Subagent requests should inherit the parent request's Auto profile by default, unless the user explicitly configures a different supported profile or model. After a response, show the effective model and profile. If a profile could not be honored because of plan, policy, capacity, or reliability constraints, provide a high-level indication of that reason.

No proprietary scoring details are needed; the goal is a predictable outcome and enough information to understand why a fallback occurred.

## Why this matters

The documentation describes Auto as selecting an optimal model based on task complexity, supported-model eligibility, system health, and availability. It does not currently give a user-facing way to choose whether Auto should optimize primarily for quality, a balanced trade-off, or economy:
https://docs.github.com/en/copilot/concepts/models/auto-model-selection

Model capabilities and prices change quickly. A routing profile would let Auto continue to evolve while giving users a stable, intentional control over the trade-off they want for coding workflows.

## Related work

- #320198 requests model inclusion/exclusion for Auto. That is complementary: an allowlist controls eligibility, while this request controls the routing objective among eligible models.
- #328630 concerns a routed-versus-executed model mismatch.
- #306717 / #306846 and #329463 cover explicit tiers or model controls for subagents/Auto.

If this is already covered by an existing issue or planned work, please consolidate it with the most appropriate item.