Scope AI generation GraphQL operations to the window's team

[warp-agent-staging]

Summary

Client-side fix for REV-2331: when the client sends no X-Warp-Team-Uid, the server's GetActiveTeamOrFallbackToPrincipal falls back to GetTeamForPrincipal, which picks the lowest team id — an unscoped, confidently-wrong answer for multi-team users. This PR scopes the three AI-generation GraphQL operations in app/src/server/server_api/ai.rs to the window's live team, using the transport support landed in #15557 (BaseClient::graphql_request_options_with_team, graphql_helpers::send_team_scoped_graphql_request).

Changes

• Added ServerApi::send_team_scoped_graphql_request, a thin wrapper around the existing transport helper that converts a RequestTeamScope into the Option<String> team uid it expects.
• Added a team_scope: RequestTeamScope parameter to the AIClient trait methods generate_commands_from_natural_language, generate_dialogue_answer, and generate_metadata_for_command, and switched their implementations to send_team_scoped_graphql_request.
• Threaded a live scope into each call site, resolved from the window/view that actually issued the request (never an account-default team):
  • AIAssistantPanelView::issue_request → Requests::issue_request → generate_dialogue_answer, scope resolved via UserWorkspaces::team_context_for_operation(ctx) in the panel's ViewContext.
  • CommandSearchView::reset_command_search_mixer resolves the scope once per mixer reset and passes it into WarpAIDataSource::new, which carries it into generate_commands_from_natural_language.
  • WorkflowModal::issue_request (Warp Drive workflow creation modal) and WorkflowView::issue_request (workflow pane) each resolve their own scope from their ViewContext before calling generate_metadata_for_command.
• None of the three operations' GraphQL variables (GenerateCommandsInput, GenerateDialogueInput, GenerateMetadataForCommandInput) name a team, so there's no risk of a header/variable mismatch.

All three call chains had a genuine live window scope available (these are user-initiated actions from an open terminal/workflow window), so nothing in this group was left unscoped.

Verification

• ./script/format — clean.
• cargo clippy -p warp --lib --tests -- -D warnings — clean.
• ./script/bundle --channel oss --nouniversal --check-only — succeeded (only pre-existing, unrelated warnings).
• cargo nextest run -p warp filtered to the touched surfaces (command_search, workflow_view, ai_assist, ai_assistant, modal_tests, warp_ai) — 26/26 passed.
• This is a plumbing change (adding a header to existing requests); no new externally-observable behavior needed a new test, and no existing tests exercised these call sites' team resolution.

[warp-agent-staging]

This PR was generated with Warp.

Comment @warp-factory on this PR to send it follow-up work.