fix: preserve provider identity during model switching

[dzianisv]

Summary

• resolve provider-backed model discovery through explicit provider aliases so name = "Azure OpenAI" still maps to the azure models.dev catalog entry even behind a proxy or localhost base URL
• preserve active_profile during config rebuild so in-session Copilot /model switching keeps the selected provider and model instead of rebuilding against the wrong provider
• harden the real TUI /model E2E mock servers to ignore partial accepted sockets so the PTY-driven model-switch flow stays stable under the real binary path
• keep the guardrail documentation added for this issue so future changes must cover the same failure modes

Testing

• cargo test -p codex-core models_dev_provider_match_accepts_azure_openai_alias -- --nocapture
• cargo test -p codex-tui rebuild_config_for_cwd_preserves_active_profile -- --nocapture
• cargo test -p codex-tui provider_model_presets_for_picker_fetches_copilot_models_for_inactive_provider -- --nocapture
• cargo test -p codex-tui --test all copilot_model_switch_then_prompt_uses_responses_api_without_cli_provider_override -- --nocapture --test-threads=1
• cargo test -p codex-tui --test all model_switching_e2e -- --nocapture --test-threads=1

Closes #21

[OpenCodeEngineer]

Pushed a725e6b1c with the actual code fix behind this PR.

Highlights:

• Azure alias handling now maps name = "Azure OpenAI" to the azure models.dev provider, including localhost/proxy base URLs.
• config rebuild preserves active_profile, which keeps Copilot model switches on the selected provider/model.
• the real /model PTY E2E mock servers now ignore partial accepted sockets instead of failing the full test run.

Local validation:

• cargo test -p codex-core models_dev_provider_match_accepts_azure_openai_alias -- --nocapture
• cargo test -p codex-tui rebuild_config_for_cwd_preserves_active_profile -- --nocapture
• cargo test -p codex-tui provider_model_presets_for_picker_fetches_copilot_models_for_inactive_provider -- --nocapture
• cargo test -p codex-tui --test all copilot_model_switch_then_prompt_uses_responses_api_without_cli_provider_override -- --nocapture --test-threads=1
• cargo test -p codex-tui --test all model_switching_e2e -- --nocapture --test-threads=1

[OpenCodeEngineer]

Remote CI on this PR is currently blocked by two separate issues outside the Rust patch itself:

• build-test is failing in npm package staging because scripts/stage_npm_packages.py cannot find a rust-release workflow for CODEX_VERSION=0.74.0. This is already tracked in fix build-test when staged npm package release workflow is missing #27.
• most rust-ci / Bazel (experimental) matrix jobs on the fork are failing before any repo steps execute because runner allocation is broken on dzianisv/codex (missing codex-runners runner group for some jobs, billing/spending-limit failure for macos-15-xlarge jobs). Tracked in Fix PR CI runner allocation on dzianisv/codex fork #29.

Examples from the current head a725e6b1c:

• Tests — ubuntu-24.04 - x86_64-unknown-linux-gnu: Required runner group 'codex-runners' not found
• Lint/Build — macos-15-xlarge - aarch64-apple-darwin: The job was not started because recent account payments have failed or your spending limit needs to be increased
• build-test: RuntimeError: Unable to find rust-release workflow for version 0.74.0.

The model/provider switching fix itself was validated locally with the targeted Rust and real /model E2E coverage listed above; the current red checks are blocked on #27 and #29 rather than a reproduced failure in the patch.

[dzianisv]

Superseded by #35, which now carries this branch's changes on the consolidated integration branch.