Record #1221 as migration phase 1 and correct its residual-race scope

PR #1221 implements the plan's phase 1 (topology-derived occupancy). Its
analysis also corrects an overstatement here: no occupancy check prevents
two sessions completing the wizard simultaneously on the same free
region, so #1220's ValueError handler remains necessary rather than
being subsumed.

Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01FsNGwvf1r6xa2JH4qiiJFf

Author: claude

docs/developer/proposals/declarative-session-reconciler-current-state.md

@@ -132,7 +132,8 @@ race today, reproducibly after #1216's gate defers builds — the grid offers
 `PlotOrchestrator.add_cell`'s overlap `ValueError`, which the success handler
 did not catch. Two fixes: catch the error (done in #1220; the simultaneous-
 wizard race needs it regardless), and derive occupancy from topology (the
-structural half, deferred to a follow-up — phase 1 of the migration plan).
+structural half — phase 1 of the migration plan, in flight as
+[#1221](https://github.com/scipp/esslivedata/pull/1221)).
 
 ## Five representations of "is anyone looking"
 

docs/developer/proposals/declarative-session-reconciler-migration.md

@@ -55,15 +55,21 @@ to this module needs.
 
 ## Phase 1 — occupancy from topology (#1219, structural half)
 
-`PlotGrid` stops deriving free positions from inserted widgets. Instead the
-reconciler hands it the occupied geometry set from the topology snapshot on
-each pass (or `PlotGrid` receives a callable). `_occupied_cells` shrinks to
-"which widget sits at which geometry" (needed for removal/replacement), and
-`_is_region_available` consults the topology-derived set.
+**Status: in flight as [#1221](https://github.com/scipp/esslivedata/pull/1221).**
+
+`PlotGrid` stops deriving free positions from inserted widgets. Instead it
+receives a callable returning the grid's topology geometries (wired to the
+orchestrator by `PlotGridTabs`), and the second click of the selection
+gesture re-validates the region before opening the wizard. `_occupied_cells`
+shrinks to "which widget sits at which geometry" (needed for
+removal/replacement) and no longer feeds selection decisions.
 
 Small, local, already agreed as the right fix in #1219. Kills the
-offered-then-refused wizard path entirely — including the cross-session race
-the `ValueError` handler now merely reports politely.
+stale-display path (empty cells offered over positions topology holds). One
+residue is unavoidable by construction: two sessions can still complete the
+wizard *simultaneously* on the same genuinely-free region, so the loser hits
+the overlap `ValueError` — #1220's error handler stays necessary as the last
+line of defence, and no occupancy model removes it.
 
 ## Phase 2 — `desired()` + differ for structure and materialization
 

docs/developer/proposals/declarative-session-reconciler.md

@@ -398,7 +398,7 @@ risks, and the test strategy are in the
 | Phase | Content | Ships alone? |
 |---|---|---|
 | 0 | Characterization tests around the current pass | yes (pure gain) |
-| 1 | Occupancy derived from topology (#1219, part 2) | yes (agreed follow-up) |
+| 1 | Occupancy derived from topology (#1219, part 2) | yes — in flight as [#1221](https://github.com/scipp/esslivedata/pull/1221) |
 | 2 | `desired()` + differ for cell existence and materialization; delete signature/version memos | yes — **this is the go/no-go spike** |
 | 3 | Fold data-flush and freshness gating into the same shape | optional |
 | 4 | Stable widget shell, swap only the figure on plotter change | speculative spike |