219 quarantined digesters are in limbo: decide whether they are deleted

[dprada]

devtools/quarantine/argdigest_argument/ holds 219 digester modules, moved there in two
batches and never decided on. They are outside the distributed package (pyproject.toml
ships molsysviewer* only), so nothing here loads, ships or resolves them — but nothing
tracks them either, which is what this issue fixes.

The directory that remains holds 362.

How they got there

First batch, 120 (uibcdf/molsysviewer#70). molsysviewer/_private/argdigest/argument/
was seeded by copying MolSysMT's wholesale. ArgDigest resolves a digester by argument name
at call time, so a digester no argument can carry is inert: imported at load, never
consulted. A name was moved only if it failed all six of: not a public argument, not a
MolSysMT attribute or alias, refused by msm.get when probed directly (0 of 120 accepted),
never consulted in a recorded full-suite run attributed by config source, referenced by no
surviving digester, mentioned nowhere in sources, tests or docs.

Second batch, 99 (uibcdf/molsysviewer#75 phase C). Delegating the get family's
digestion to MolSysMT removed the only surface that kept our copies of their attribute
digesters reachable. With it gone they answer to nothing.

Both batches were verified the same way: STRICTNESS temporarily "error", so a missing
digester raises instead of warning, then the full suite. Zero No digester for in both.

The decision that is open

Delete them, or keep them. Nothing else needs doing — they are correct where they are.

For deleting: they cannot run. Keeping unreachable copies of another package's code is
precisely what uibcdf/molsysviewer#70 is about, and 219 files is a third of what that
directory once held.

For waiting: the smoke test in devguide/smoke_test.md exercises paths the automated
suite does not, and it has not been run since the second batch moved. If any of these is
reachable only from a flow a human drives, that is where it would show.

Against deleting quietly: the six tests are heuristics stacked to be conservative, not a
proof. My earlier estimate of this same set was wrong twice — first 273 files by a method
that missed the **kwargs surface entirely, then a spot-check that re-ran the same flawed
test and "confirmed" it. The measurements that stand are the ones taken after those
corrections.

Recommendation

Delete after the manual smoke test passes, not before. git mv restores any of them, and
devtools/quarantine/README.md carries the evidence for each batch and the restore command.

Related: uibcdf/molsysviewer#70 (why the duplication exists at all), #75 (phase C).