refactor(ssr): Phase 4 — full Selmer migration of POS Sales Summary; remove the wizard; fix add-item + totals

Migrates the POS Sales Summary edit modal off the wizard to a plain Selmer form,
building on the parity gate committed earlier. Largest migration so far and the
first with no prior test coverage.

What changed
- Wizard removed: deleted MainStep/EditWizard records, MultiStepFormState, the
  step-params[...] prefix, the EDN snapshot round-trip, and all mm/* middleware.
  Replaced with a plain handler + flat wrap-decode/wrap-derive-state. The 51 fc/
  cursor refs are de-cursored into explicit data + Selmer templates.
- db/id-keyed item merge: wrap-derive-state overlays posted items onto the
  persisted items by :db/id, so read-only fields the form doesn't post
  (ledger-side, amount) survive a re-render and the debit/credit split + totals
  stay correct. New manual rows (temp db/id) ride through as-is.
- Inline click-to-edit account cell preserved as three small targeted
  .account-cell-swap routes (edit/save/cancel-item-account), ported to Selmer
  with the new field-name scheme.
- 100% Selmer modal render path (the remaining Hiccup / hx-swap-oob / "hx-"
  strings are all grid-page code — grid render lambdas, the filters form, and the
  submit response-header map — not the modal).
- Routes: dropped edit-wizard-navigate + new-summary-item; added form-changed.

Fixes (two pre-existing bugs, per request)
- "New Summary Item" add button (was throwing `newRowIndex is not defined` and
  targeting a non-existent `.new-row`) is now a whole-form-swap op=new-item that
  adds an editable manual row (category + account typeahead + debit/credit money
  inputs + remove).
- The dead totals/balance display (malformed Hiccup that discarded its labels) is
  replaced by a proper #summary-totals block showing running Total +
  Balanced/Unbalanced, refreshed via a Rule-4 targeted swap on manual amount edits.

Scorecard delta (pos/sales_summaries.clj): LOC 790->732, mm coupling 20->0,
wizard records 4->0, fc/ cursor 51->0, step-params 27->0 (2 comments), modal
routes 8->6. (hx-swap-oob 1 and mixed-hx live in the grid page, not the modal.)

Verification: sales-summary spec 7/7 (incl. the two fixes); full Playwright suite
46/46; cljfmt clean. Skill fed: scorecard row + narrative; gotchas (parity-gate-
first, characterize-then-fix, keyup-trigger tests); cookbook (inline click-to-edit
cell, db/id-keyed item merge).

Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>

.claude/skills/ssr-form-migration/reference/gotchas.md

@@ -224,6 +224,35 @@ on each post (`[:vector {:coerce? true} entity-id]` + the `coerce-vector` transf
 — you code exactly the rows the user saw, immune to data changing between open and submit. This
 is heuristic 2 → 0 for a multi-select modal.
 
+## No parity gate? Build one first — seed + characterization spec, before touching code
+
+A modal with **no e2e coverage** (and no test-server seed for its domain) cannot be migrated
+safely — "behavior parity is proven by tests, not by reading" is the skill's #1 non-negotiable.
+Phase 4 (POS Sales Summary) had zero coverage. The fix: (1) seed a representative entity in
+`test_server.clj`'s `seed-test-data` and surface its id via `/test-info`; (2) write a
+characterization spec against the **unmodified** modal and confirm it green; (3) commit the gate
+*separately, ahead of the rewrite*. Reach the modal the real way (grid → row's edit button), not
+a direct fragment URL. To discover the actual rendered structure (field names, ids, swap targets)
+— especially when the code has dead/buggy render fns — dump the live modal HTML with a throwaway
+spec first; assert against what *renders*, not what the code looks like.
+
+## Characterize before you fix; never assert a bug as working
+
+Writing the gate often surfaces pre-existing bugs (Phase 4: a "New Summary Item" button that
+threw `newRowIndex is not defined`, and a totals display whose malformed Hiccup discarded its
+own labels). Do **not** assert the broken behavior as if it works, and do **not** silently "fix"
+it mid-refactor — surface it and let the user decide fix-vs-preserve. If they choose *fix*: the
+spec first documents the break (a passing test of the *current* inert behavior or an explicit
+note), then is rewritten to assert the *fixed* behavior as part of the migration commit.
+
+## htmx `keyup`-triggered inputs need real keystrokes in tests
+
+A money/text input wired `hx-trigger="keyup changed delay:300ms"` does **not** fire on Playwright
+`.fill()` + `dispatchEvent('change')` — `fill` sets the value without keyup events. Use
+`.click()` then `.pressSequentially('500')` (types char-by-char, firing keyup) so the targeted
+swap actually triggers. (A `change`-triggered control is the opposite — `dispatchEvent('change')`
+is fine there.)
+
 ## Scorecard exceptions (ratchet violations with a reason)
 
 **Heuristic 4 (LOC net ↓) — exception (Phase 3, Transaction Bulk Code: 420→506).** When the