Gate the four change-directing fields on there being a change

[vijay-tylertech]

BP updates, Fix target, KB action and KB files DESCRIBE work that prose has asked
for. Set on their own they described a change nobody had described, and BP updates
was the sharp end: ticked with no prose it opens a Blueprint change request carrying
no statement of what should be different.

So all four moved out of the lower grid and sit directly under Overall suggestions
and comments -- beside the box that has to be written first, which is the shortest
way to say that is the order -- under the heading "Directs the change", disabled and
dimmed until an Ideal response or a Summary exists. BP updates is unchecked by
default. They are still NOT triggers: prose remains the only trigger, per
wants_change(). These four direct the work it asks for.

THE GATE IS NOT PURELY "IS THERE PROSE", deliberately. It also opens when the fields
already hold values, because otherwise merely OPENING a transcript that recorded them
without prose -- which legacy transcripts do -- would clear a reviewer's work on page
load. Silent data loss triggered by reading a page is worse than a stale field. The
gate therefore closes only while everything is empty, which is the fresh-transcript
case the rule exists for, and "No changes & next" clears all four explicitly: a
transcript recorded as no-changes must not carry a Fix target, a KB action, KB files
or a ticked BP updates, or the batch would try to act on it.

Also here:

• The SAVE BADGE counts both repos. It counted this working tree only, so a batch of
  3 transcript edits and 2 staged Blueprint files put "3" on the nav beside a line
  reading "5 edited file(s) not yet saved" -- two numbers for one thing, and the
  smaller one is what is seen from every other page. The JS that patches the badge
  after an action already sent the combined figure, so the two paths now agree.
• "Add an Ideal response under each Exchange" -- capital I, matching the field label.

Verified in Chrome across four states: empty (all four disabled, dimmed, hint shown,
"No changes & next"); an Ideal response written (all enabled, hint hidden, "Changes
suggested & next"); a Summary alone (same); prose removed again (all four disabled and
reset to none/unchecked/Select...). Badge and status line both read 5 against 3 local
plus 2 Blueprint. Contrast green.

Claude-Session: https://claude.ai/code/session_01VVNYqGza2dFdeWFX7qXnwb