🤖 fix: enable GPT-6 Astra Pro mode through Coder - #4124

Allow Pro mode on Coder's OpenAI Responses instances without broadening the
Fast mode or incompatible gateway gates. Keep raw instance metadata authoritative
and verify the picker, persistence, and native SDK request serialization.

---
_Generated with [`mux`](https://github.com/coder/mux) • Model: `coder:openai/gpt-6-astra` • Thinking: `high` • Cost: `$20.61`_

<!-- mux-attribution: model=coder:openai/gpt-6-astra thinking=high costs=20.61 -->

Resolve mapped Pro capabilities, preserve type-derived identities on direct
fallback, and honor catalog exclusions/removals before restoring Coder routes.
Pin the phone story's container to Pixel's width and update advisor coverage for
the newly supported Coder route.

---
_Generated with [`mux`](https://github.com/coder/mux) • Model: `coder:openai/gpt-6-astra` • Thinking: `high` • Cost: `$36.93`_

<!-- mux-attribution: model=coder:openai/gpt-6-astra thinking=high costs=36.93 -->

Honor policy-aware explicit gateway fallback in the picker and palette without
letting Treat-as metadata affect routing. Preserve scoped Pro capabilities and
actual Coder instance identity through nested advisor option construction.

Validated with 440 targeted tests, make static-check, both Astra Pro Storybook
plays, and recorded phone/desktop browser checks.

---

_Generated with [`mux`](https://github.com/coder/mux) • Model: `coder:openai/gpt-6-astra` • Thinking: `high` • Cost: `$6.17`_

<!-- mux-attribution: model=coder:openai/gpt-6-astra thinking=high costs=6.17 -->

Project nested provider options from the same raw snapshot used for model
creation so catalog refreshes cannot change the instance wire or scoped aliases.
Cover retyped and removed metadata, update routing-hook test doubles, and restore
routing/minimum hooks so the Settings suites do not leak those mocks.

Validation: 587 combined UI/provider tests and make static-check on Bun 1.3.5.

---

_Generated with [`mux`](https://github.com/coder/mux) • Model: `coder:openai/gpt-6-astra` • Thinking: `high` • Cost: `$42.99`_

<!-- mux-attribution: model=coder:openai/gpt-6-astra thinking=high costs=42.99 -->

Create one request receipt with effective route, wire, scoped metadata and
provider options. Continuous summaries and nested tools consume that receipt
instead of rebuilding routing from a raw prefix or live configuration.
Preserve unset-thinking variant behavior and leave metadata-only callers alone.

Cover Coder aliases, unavailable/catalog fallbacks, protocol overrides,
configuration/route races and nested advisor behavior. Validated with 891
combined tests and make static-check on Bun 1.3.5.

---

_Generated with [`mux`](https://github.com/coder/mux) • Model: `coder:openai/gpt-6-astra` • Thinking: `high` • Cost: `$52.52`_

<!-- mux-attribution: model=coder:openai/gpt-6-astra thinking=high costs=52.52 -->

Extend the existing disabled Coder shell with parsed instance names/types so
persisted selections can resolve allowed upstream fallback capabilities.
Keep forbidden Coder unconfigured, unselectable and absent from provider lists;
expose no model catalog, endpoint settings or authentication data.

Preserve credential-presence revocation and custom-provider shadow semantics.
Validated with 895 combined tests and make static-check on Bun 1.3.5.

---

_Generated with [`mux`](https://github.com/coder/mux) • Model: `coder:openai/gpt-6-astra` • Thinking: `high` • Cost: `$73.91`_

<!-- mux-attribution: model=coder:openai/gpt-6-astra thinking=high costs=73.91 -->

Spy on the summary model's migrated creation method and return its full
request receipt. Preserve plugin policy, cancellation, cleanup and usage
assertions; leave metadata-only callers and production code unchanged.

Validated build-main, all 13 isolated CI suites (440 tests), and static checks.

---

_Generated with `xum` • Model: `coder:openai/gpt-6-astra` • Thinking: `high` • Cost: `$92.28`_

<!-- mux-attribution: model=coder:openai/gpt-6-astra thinking=high costs=92.28 -->