Feedback survey: follow-up text results and a narrower every_rating

What

After you rate a response, Claude Code can show a feedback survey with a box for more comments. A host is an app that runs Claude Code inside it, such as an editor. Several pieces of this changed:

• Follow-up details: the survey record a host sends now carries followup_box, no_followup_reason, char_count, error_code and saved_locally.
• Follow-up events: followup_text_submitted and followup_text_failed are now accepted, and are handled before the usual survey check.
• logs_followup: a new field tells hosts whether this version logs the followup_* survey events they send. The feedback and transcript-sharing information now always includes logs_followup: true.
• every_rating: this value, which marks every rating for collection, now needs one more value to be present on top of the tengu_curried_sparrow server switch.

The previous release built tengu_curried_sparrow as off by default, and no reading has been taken under this release yet.

Why

An editor's feedback box can now report whether follow-up text was sent, saved or failed. Every rating is marked for collection in fewer cases than before.