emrg: a bookmark row stamped later than the clock reads every project as unchanged, for ever #2421

What is missing

scripts/check-memory-rotation.py answers §1.5's step 0 — for each registered project, is its memory newer than the bookmark row that records what has already been read?

project_state() parses a row's stamp with datetime.strptime(value, STAMP_FORMAT) and hands it to newest_memory(root, int(stamp.timestamp())), which keeps the files strictly newer than that instant. Nothing compares the stamp against the clock. A stamp later than now therefore keeps nothing: every project reads unchanged, and once every project has a row the report prints its clean line and exits 0:

changed since their bookmark: 0 - every project is at or before its stamp
states: unchanged 2
coverage: every project has a row and every row names a project (2 of 2)

A row stamped in the future cannot describe anything a cycle has read, so that line is a confident clean verdict about a file the tool cannot believe — and it does not change by itself, because the file stays in the future until someone edits it. It is the same silently-forever failure the tool's own docstring catalogues twice: the ISO T form answers 0 files for ever, and a sub-second comparison reports the very file a row was written from as newer for ever.

Measured on this host

A two-project register with one memory file each, and two bookmarks differing only in the stamps their rows carry:

Both runs pass the tool's argv (--projects, --bookmark); neither writes anything.

What a fix must satisfy

1. A row whose stamp is later than the instant of the reading is refused as unmeasurable — exit 2 — naming the project, the stamp it carries, and the instant it was compared against, so that a clock-skew cause is visible rather than assumed. It belongs beside the existing refusal in the same function, the one for a stamp that does not parse at all.
2. A row stamped at or before that instant keeps today's reading, unchanged included: a test in both directions, because these two cases differ by a second and by nothing else.
3. The refusal is reachable from the tool's argv, not only from a unit call.
4. The tests take their constants from the tool (STAMP_FORMAT, BASE-style fixtures already in the file) rather than writing the format a second time.