fix: move the heart to week 37, where it can still be drawn

[vyncint]

The heart could not be drawn. Not "behind" — holed, mossaic's word for a
plan where days inside the picture are already brighter than the picture
wants them, which no amount of future contributing takes back.

Heart · 2026 · week 35, 11 columns — tracking vyncint
  Cannot be drawn cleanly — 5 of 56 shaded days are already too bright.

Five days, four of them on days the heart needs dark:

day	has	wants
2026-08-30	7	at most 0
2026-09-02	82	at most 70
2026-09-03	41	at most 0
2026-09-04	2	at most 0
2026-09-05	36	at most 0

Why week 35 stopped working

The README's reason for choosing 35 over 34 was that it was "the first
column still entirely in the future". It was, when the plan was made. Week
35 began on Sunday 30 August, and between then and now ordinary work landed
inside it. The rule was right; the placement simply aged out from under it.

Where it goes

Swept every placement that fits the year, against live data today:

start week	verdict	holes
33	holed	13
34	holed	8
35	holed	5
36	holed	1
37 – 42	reachable	0
43+	overhangs the year	—

37 through 42 are all clean and all cost exactly the same — 56 days at 36
contributions each, 2,016 in total. Week 37 is the earliest, begins Sunday
13 September so it is still entirely in the future (the original rule, held
to), and keeps the most slack: the last lit day is 25 November, thirty-six
days before the year ends. Week 42 is the last that fits.

After:

Heart · 2026 · week 37, 11 columns — tracking vyncint
  On track — 0 of 56 shaded days are right, 56 to go.
  level  days   done   owing   each
      2    56      0   2,016   36
      0    21     21       -   must stay dark
  still owing  56 days · 2,016 contributions
  today        outside the picture — nothing owed

The rule this edit breaks

"Keep this fixed: the plan is these inputs, and changing one compares
against a different plan." That rule is the reason this PR carries a
paragraph in the workflow and a paragraph in the README saying what moved,
when, and why — the numbers in reports from before today are not comparable
with the ones after it. This restarts the plan on purpose.

Three corrections carried in the same pass

All measured, not inherited:

• 36 per day, 2,016 total — the README said 35 and 1,960. The year's
  peak has moved since it was written, and the peak is what prices a shade.
• eleven columns — the "Why a picture" paragraph said thirteen. The
  .art file, the table and the tool all say eleven.
• the placement guard now names where the placement actually shows up.
  mossaic 0.7.0 puts it in the report header (· week 37, 11 columns) and
  exposes it as the start-week and columns outputs; before that it
  appeared on no line at all, so the guard pointed at nothing.

[vyncint]

Rechecked before merging, and week 37 had aged out the same way week 35 did:

$ mossaic-art --matrix heart.art --year 2026 --start-week 37 --track vyncint
  Cannot be drawn cleanly — 4 days are brighter than the picture wants,
  and nothing takes a contribution away.
  --start-week 39 draws it cleanly.

Week 38 is holed by one day; 39 is the first clean one today. Merging this as written would have reintroduced exactly the bug it was opened to fix — this branch was correct on 6 September and sat for fifteen days.

Pushed f59022a moving it to 39 (Sunday 27 September → 12 December, last lit day 9 December, twenty-two days of slack instead of thirty-six — the price of the delay). README and track.yml both updated, and the history in each now records both moves with the same cause.

The part worth keeping: the rule was never wrong. "The first column still entirely in the future" is correct at the moment you measure it, and the gap between choosing a placement and landing it is what breaks it. So track.yml's comment now says to measure on the day it lands and names the command that prints the earliest clean column.