Keep exported CSV cells on one line so Excel renders normal rows

[vijay-tylertech]

NOT a parsing problem -- the file parses as exactly 4 columns and always did; the
empty columns to the right of D in the report are Excel's own, not data. The defect
is row HEIGHT: a markdown answer carries 20-30 newlines (26 and 17 in the two rows
reported), a CSV cell may legally contain them, and Excel wraps such a cell and
grows the row to fit. A two-row export therefore rendered as two screen-height
blocks with the text in a column a few characters wide.

Nothing in the format can fix this -- a .csv carries no column widths and no row
heights -- so the only lever is the content. The two reference columns are now
collapsed onto one line, with paragraph breaks becoming a pilcrow rather than
disappearing: the reader has to judge an answer from that cell, and dropping every
structural boundary would turn 1,400 characters into an unbroken wall.

The Correction column is deliberately NOT touched. It is the one a reviewer authors,
and re-flowing their line breaks on each export would edit their prose a little more
every round trip.

Matching is unaffected: _qkey collapses whitespace on both sides, so a question
re-flowed on export still matches the transcript it came from.

Verified: one physical line per logical row, zero newlines in every cell, pilcrows
preserved (8 and 6), _qkey equal across the reflow, a full export -> fill column D
-> import cycle landing both corrections with nothing ignored, and a multi-line
correction surviving re-export byte-exact while the answer beside it stays
single-line.

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