Refactor context module: extract utilities, consolidate error handling

[alexkroman]

Summary

This PR refactors the aai_cli/context.py module to improve code organization and reduce duplication. The changes extract reusable utilities into dedicated modules, consolidate error-handling patterns, and remove thin adapter functions that are no longer needed.

Key Changes

• Extract stdio utilities: Move interactive_stdio() from main.py to a new aai_cli/stdio.py module (alongside existing piped_stdin_text() and stdin_is_piped()), making the "is this a real terminal?" predicate reusable across the codebase.

• Extract subprocess spawning: Create aai_cli/procs.py with spawn_detached() to centralize the detached-subprocess recipe used by both telemetry and update-check, eliminating duplication and ensuring consistency.

• Consolidate error handling: Introduce _fail() helper in context.py to unify the error-emit-and-exit pattern used throughout run_command() and _auto_login_and_exit(), replacing repeated output.emit_error() + typer.Exit() sequences.

• Lazy-load auth module: Move the from aai_cli.auth import run_login_flow import inside persist_browser_login() to avoid importing the auth package (httpx, Stytch, loopback server) on every command's import path—only when a browser login actually starts.

• Remove thin adapters: Delete module-level resolve_profile(), resolve_environment(), resolve_session(), and env_override_warning() functions that merely delegated to AppState methods; callers now invoke the methods directly.

• Simplify speak_exec.py: Use the new stdio.piped_stdin_text() helper and extract a _emit() function to deduplicate JSON/human output logic in _emit_single() and _emit_multi().

• Update main.py: Replace the hand-maintained _ROOT_ONLY_FLAGS list with dynamic flag discovery from the root callback's parameter declarations, ensuring new global flags automatically get placement hints without manual updates.

• Refactor llm.py: Simplify _list_models() by removing the higher-order function wrapper and inlining the body directly into the command handler.

• Extract render_check_lines(): Move the doctor-check rendering logic to aai_cli/commands/doctor.py so the onboarding wizard can reuse it without duplication.

• Update test mocks: Adjust all test patches to target aai_cli.auth.run_login_flow (the new import location) instead of aai_cli.context.run_login_flow.

• Consolidate error factories: Extract _no_project_error() in auth/flow.py to avoid repeating the same error message in two branches.

• Simplify config validation: Remove the private _validate_profile() wrapper; callers now use validate_profile() directly.

Notable Implementation Details

• The _fail() helper uses from None to suppress the exception chain for user-facing errors (keeping output clean), but the internal-error path in run_command() preserves the chain with from exc for debugging.

• stdio.interactive_stdio() is now the single source of truth for "can we prompt the user?" across the setup wizard, onboarding prompter, and assembly init template picker.

• The procs.spawn_detached() recipe ensures detached children have their own session, discarded stdio, and a self-disabling env var to prevent recursive spawning.

• argscan.JSON_FLAGS is now a module-level constant shared with main.py's misplaced-flag hint logic, preventing drift between the two.

https://claude.ai/code/session_01APHW6krqHMsPQ7MsMuBR9n