feat(assistant-editor): tailor chat input controls for the preview

[philmerrell]

Summary

• Hide voice mode and the settings button in the assistant editor preview chat input — neither is wired into the editor — and expose file attachments so authors can test their assistant against the same inputs end users will send.
• Wire fileUploadIds from the preview chat input through PreviewChatService to /chat/stream as file_upload_ids (mirrors the field name used by the main chat).
• Split chat-input's settings button into its own showSettingsControl gate (previously coupled to showFileControls) and add a parallel showVoiceControl. Both threaded through ChatContainerConfig so all chat-input call sites can opt in/out independently.
• Rename chat-input's hidden file <input> id to chat-input-file-upload to avoid colliding with the assistant form's existing knowledge-base file input (both previously used id="file-upload", which became a duplicate on the editor page once attachments were re-enabled in the preview).

Notable choices

• KB and chat-input accept lists remain separate by intent: chat attachments include images and no .pptx; the KB upload includes .pptx and no images. Unifying them is a separate decision.
• PreviewChatService.sendMessage gains a 4th optional fileUploadIds arg; the request body only adds file_upload_ids when files are present.

Test plan

• Open an assistant in the editor — preview chat input shows the + attach button but no mic and no settings/sliders.
• Attach an allowed file in the preview, send a question that references it, confirm the assistant response reflects the file contents (proves file_upload_ids made it through).
• On /session (new chat), confirm all four controls remain visible: attach, settings, mic, send.
• On the editor page, click the KB "Add content" upload and the preview's + separately — each opens with its own (different) accept list, no duplicate-id console warnings.

🤖 Generated with Claude Code