TASK-122: a bullet slot is not a cell, so render --write invents no padding

[ranjiao]

The bug

bin/perry_store.py § describe_cell's disagreement branch defaulted the
padding to one space on each side. For a table cell that is right —
render_line joins on |, a character carrying no whitespace of its own. For
a bullet slot it is duplication: slot_descriptor's literal span already
holds the separator.

input           '- Repo layout: single'
literal span    '- Repo layout: '        ← the space after the colon is already here
lead / trail    ' ' / ' '                ← invented on top of it
rendered        '- Repo layout:  split '

Two spaces after the colon, and a trailing one git diff --check reports.
bin/perry_md_store.py:972 prints, in its own refusal message, `<tool> render --write` to bring the file back in line, so the tool was
instructing the reader into the damage.

The rule

Stated once, where the padding is decided, keyed on the same escape flag that
already distinguishes a cell from a slot: only a table cell may be handed
padding it did not come with. The whitespace-only branch follows the same
rule — a bullet keeps that whitespace whole and on the left, where the
separator it follows is.

Verification (V3)

1. - Repo layout: single + a store holding split renders
   - Repo layout: split, byte-exact.
2. Reverting the rule reddens the three bullet cases and NOT the table-cell
   case (test_a_table_cell_that_lost_its_padding_is_still_given_it_back
   stays green, as do test_a_track_row and the blank-marker test). The two
   paths are genuinely separated.
3. test_the_declared_blank_marker_survives_the_bullet_path green, unedited.
4. Run, not asserted — a real .perry/config.md in a temp git repo, drifted,
   perry-config render --write, then git diff --check: silent, exit 0. Also
   covered as a test.
5. python3 tests/parallel -j 4: 65 modules · 1934 tests, one pre-existing red
   module (test_diagnose, 2 failures about dangling DESIGN-900 / REL-00 /
   ZZZ-404) identical to the baseline measured on this branch point.
   bash tests/run agrees. python3 bin/perry-lint 0 errors.
   git diff --check clean. git diff -- perry/ empty.

Diff size

Own code diff: 2 files, +42 / −7. The PR's file count is inflated because
feat/work-modes is ahead of origin.

🤖 Generated with Claude Code