Shows the drafts tray in the reference embedder

[johnnyt]

The drafts tray, visible in the reference embedder, and the beat it exists
for: a flow authored backwards from its sink.

The interim pin

statifier_blocks moves from Hex ~> 0.11 to a git ref on that repo's
main, a3833479257fb0692eea65ed50c00c939b489f36 - the merge of its PR 201,
the first commit carrying core.drafts, core.placeholder,
StatifierBlocks.Shelf and the .sb-slot--tray strip. 0.11.0 carries none
of them, so the fixture below would name two types no palette resolves.

This is INTERIM, on the established se-p22 pattern. FINAL re-pin
after the operator publishes statifier_blocks 0.12.0: back to
{:statifier_blocks, "~> 0.12"}, retiring @statifier_blocks_ref and its
guard with it. Tracked by se-l9s's third half, alongside the open
statifier and statifier_persistence halves.

mix.lock moved by exactly that one entry. test/statifier_examples/ mix_deps_test.exs guards it the way the other two pins are guarded: one
@statifier_blocks_ref asserted against both mix.exs and mix.lock, so
an edit to one without the other cannot pass.

The uptake needed no wiring, and that is the finding

Nothing in lib/statifier_examples_web/** changed. The two new types reach
this app's editor on their own:

• StatifierExamples.Charts.palette/0 builds with
  Palette.from_modules(registrations(), core: true), and core_types/0
  is now 15 rather than 13, so the palette is 28 entries rather than 26;
• assets/css/app.css already imports the package stylesheet by path, and
  .sb-slot--tray is in it - no new --sb-* token, so no host theme
  change;
• core.drafts declares the icon "inbox", and this app resolves icon
  names against the whole heroicons 24/outline set rather than a
  hand-written subset, so the tile draws.

For a package whose uptake bead expected palette registration and CSS
passthrough, that is worth recording as the reference embedder's answer:
the surface arrived and the host had nothing to do.

What the app does add

priv/fixtures/card_processing_sketch.json (bdoc_cp_sketch, key
card_processing_sketch) - the same payment flow as card_processing,
caught halfway through being authored. The tail the author already knows,
a Sequence of Build the receipt and Send the receipt, is parked
in a core.drafts shelf at the root's body; a core.placeholder stands
where the middle will go; a core.assign seeds capture.attempts; and the
datamodel declares amount and authorization ahead of the steps that
will use them, which is what authoring from the sink backwards looks like
in the document. It opens at Findings 2 - a shelf with anything in it and
a placeholder each raise an author warning on a compile that succeeds.

docs/demo-script.md gains section 13, the sink-backwards beat, and its
one falsified row is corrected (beat 1's palette count, 26 -> 28).

StatifierExamplesWeb.EditorLiveTest gains the beat's machine-verified
half: three rows that mount the sketch, assert the tray draws with the tail
in it, and then drive the whole sequence through the editor's own
insert-dragstart / insert-drop / remove / dragstart / drop
events - park the tail, build the sources, place the tail - ending with an
empty tray, the tail in the flow, and the header at Findings 0. That zero
is the publish gate read off the page: what a host does about parked work
is host policy off %Compiled{}.warnings, and this is that policy walked.

Four sibling assertions the new fixture falsifies are updated in place: the
fixture list in charts_test, the registration row in card_auth_test,
the per-fixture label counts in step_label_test, and the declared-roots
list in fixture_datamodel_test.

Evidence

Live on a private port (8651) on this branch, through the promoted DevTools
driver - the frozen demo ports were not touched:

• se-ihm-light-tray.png, se-ihm-dark-tray.png, se-ihm-brand-tray.png -
  the tray at the foot of the canvas with the tail parked in it, in all
  three themes;
• se-ihm-light-tray-drag.png - a palette entry dragged over the tray,
  driven by DragEvents carrying a DataTransfer from page JS
  (se-ihm-drag.js); the tray stamps data-drop="ok", which is
  palette-to-tray drop being permitted by the declaration rather than by
  new machinery;
• se-ihm-light-tray-placed.png - the settled end state after the same
  sequence the test drives, run in a real browser (se-ihm-beat.js):
  Findings 2 -> Findings 0, 8 blocks/depth 4 -> 9 blocks/depth 3, tray empty.

Every mutation was run by hand and shown red before being reverted from a
backup copy: the shelf retyped to core.sequence, the shelf's body
emptied, the pin ref pointed at the previous statifier_blocks main
487cebf1, and the sketch dropped from @documents.

Full mix quality green on the branch: 243 tests, 77.5% coverage, dialyzer
clean.

Provenance

• Bead: se-ihm (campaign 024, Lane A uptake A3).
• Upstream: statifier_blocks PR 201 (sb-uag7), whose closing notes are
  the facts this uptake is written against.
• Implements the editor half of campaign-024 ruling R-b (a shelf is a
  direct child of the root's body, at most one per document) as a
  consumer of it; the rule itself is enforced upstream.
• No docs/adr/ path in this diff, so no direction-agent review gate
  applies.
• No sibling PR was rebased over: the branch is cut from main at
  5bafc2f2 and no 024 sibling has touched these files.