feat: ADR-0035 Amendment 1 — drop composite UNIQUE; friendlier drop-column + generic-error wording

F1/F2/F3 from the whole-Phase-4 /runda (handoff-42 §3):

- F3: drop an anonymous composite UNIQUE via a derived, engine-neutral
  name `unique_<cols>` — recomputed live, nothing persisted, reusing the
  existing `DROP CONSTRAINT <name>` grammar (no new syntax/metadata, the
  §4g anonymity decision intact). A name matching more than one UNIQUE is
  refused as ambiguous, never guessed. One undo step. `describe`
  annotates each composite UNIQUE with its name.
- F1: dropping a column a composite UNIQUE covers is refused up-front
  with the derived name + the actionable drop command (was an unhelpful
  generic engine refusal).
- F2: contextless friendly_message() no longer leaks a literal `{table}`
  in the generic hint (new `error.generic.hint_no_table`, selected when
  no table is in context). The table-ful path is unchanged.

Docs: ADR-0035 Amendment 1 + Status + README index + plan
docs/plans/20260526-adr-0035-composite-unique-drop-f1f2f3.md.
Tests: +5 (drop-by-name, ambiguous-refused, one-undo-step, F1 guard,
F2 no-leak) + a describe-render assertion. 1922 pass / 0 fail / 0 skip;
clippy clean.

docs/plans/20260526-adr-0035-composite-unique-drop-f1f2f3.md

@@ -0,0 +1,110 @@
+# Plan — ADR-0035 Phase-4 `/runda` follow-ups F1 / F2 / F3 (2026-05-26)
+
+Bundle of the three error-message / capability follow-ups surfaced by the
+whole-Phase-4 `/runda` (handoff-42 §3). All three live on the **safe**
+composite-UNIQUE edge (dropping a UNIQUE-covered column is correctly
+*refused* today — no corruption); the work improves messaging and adds a
+way to drop the constraint itself.
+
+## Phase 1 — requirements
+
+- **F1 — friendly refusal for dropping a composite-UNIQUE column.**
+  `do_drop_column`'s covering-index guard reads `read_table_indexes`,
+  which filters to `origin='c'` (explicit `CREATE INDEX`) and excludes
+  the UNIQUE-constraint auto-index (`origin='u'`). So `drop column c`
+  when `unique (b, c)` spans `c` skips the guard, reaches the engine, and
+  is refused with an unhelpful generic message. Add an up-front guard
+  detecting the column in `schema.unique_constraints` (composite only —
+  `read_unique_constraints` routes single-column UNIQUEs to the column
+  flag, multi-column to `unique_constraints`), refusing with the
+  constraint's **derived name** (F3) + the drop command. Behaviour stays
+  "refused"; only the message improves. **Message-only — no `--cascade`
+  extension** (the SQL drop-column has no `--cascade` spelling; dropping
+  a constraint via cascade is a larger semantic change, out of scope
+  unless the user asks).
+
+- **F2 — literal `{table}` leak in contextless `friendly_message()`.**
+  `Verbosity` defaults to `Verbose`, so `friendly_message()` (which uses
+  `TranslateContext::default()`, no table) renders the generic hint
+  `"…current state of `{table}`."` with the literal placeholder via
+  `ctx_table()`'s `"{table}"` fallback. Hits every contextless
+  `friendly_message()` callsite whose error lands in the **generic
+  bucket**: replay, undo, rebuild-from-text, export. Fix: a tableless
+  generic-hint variant selected when `ctx.table` is `None`. **Broader
+  finding** (DA): the same `{name}`-marker fallbacks leak in *other*
+  templates (e.g. a replayed UNIQUE violation → `error.unique.*`) when
+  reached contextless. The documented F2 is the generic case; the broader
+  leak is surfaced for the user to scope, not silently expanded/narrowed.
+
+- **F3 — a way to drop an anonymous composite UNIQUE (user-raised).**
+  By design (§4a.2/§4g) a composite `UNIQUE(a,b)` is anonymous —
+  PRAGMA-detected, a bare column-list, no name — so `DROP CONSTRAINT
+  <name>` can't target it and recreating the table is the only escape.
+  Add a way to drop it. **(Amends ADR-0035 — see Amendment 1.)**
+
+Baseline: `cargo test` → 1917 pass / 0 fail / 0 skip / 1 ignored doctest;
+`cargo clippy --all-targets -- -D warnings` clean.
+
+## Phase 2/3 — F3 design (the genuine fork; user-decided)
+
+Composite UNIQUE has no name. Options considered:
+
+- **A — name composite UNIQUEs (user-supplied):** reverse the §4g
+  anonymity decision; needs a new `__rdbms_*` table + YAML round-trip +
+  rebuild-arrival migration (the cost §4a.3 deliberately avoided). Most
+  SQL-standard, largest.
+- **B — positional drop by column-list** (`drop … unique (cols)`):
+  preserves anonymity, no metadata, but needs a *new* grammar form.
+- **C — auto-assigned, engine-neutral *derived* name (chosen).** The
+  name is a deterministic function of the columns (`unique_<cols>`),
+  recomputed live wherever shown or matched. Storage stays a bare
+  column-list (anonymity preserved); the name is purely a
+  presentation/addressing label. **Reuses the existing `DROP CONSTRAINT
+  <name>` grammar — no new syntax at all.** Zero metadata, zero
+  migration, round-trips for free. Tracks column renames.
+
+**User decisions (2026-05-26):** approach **C / derived (no storage)**;
+name format **`unique_<cols>`**; doc vehicle **amend ADR-0035**; scope
+**advanced-SQL only** (matching the 4g `ADD` form — no simple-mode verb).
+
+### DA critique (written down)
+
+1. **Ambiguous derived names** (e.g. a column literally named `b_c` vs
+   `UNIQUE (b, c)`): drop-by-name must **detect ambiguity and refuse**,
+   never guess. *In scope.*
+2. **Collision with a user-named CHECK/FK** of the same string: the
+   `do_drop_constraint_by_name` order is CHECK → FK → UNIQUE, so a
+   CHECK/FK shadows a derived UNIQUE name. Acceptable given the
+   distinctive `unique_` prefix; **document the order**.
+3. **F1 `--cascade`**: not extended to drop a covering UNIQUE
+   (constraint, not index). Refuse-only. *Flagged.*
+4. **F2 breadth**: the leak is broader than `error.generic.hint`. Fix the
+   documented generic case; **surface** the broader leak. *Flagged.*
+5. **Single-column UNIQUE column drop**: a parallel gap (a single-column
+   UNIQUE column drop also reaches the engine with a poor message) exists
+   but is **outside the documented F1 scope** (different mechanism —
+   ADR-0029 column-level `drop constraint`). Noted, not fixed here.
+
+## Phase 4 — execution (order: F3 → F1 → F2)
+
+1. **F3.** `unique_constraint_name(cols) -> "unique_<cols>"` helper
+   (`db.rs`, `pub(crate)`). Extend `do_drop_constraint_by_name` with a
+   third step: match each composite UNIQUE's derived name; >1 match →
+   refuse (ambiguous); 1 match → `rebuild_table` with that entry removed
+   from `unique_constraints` (mirrors `do_alter_add_unique` in reverse) +
+   `do_describe_table`. Annotate the describe "Table constraints:"
+   section: `unique_b_c: UNIQUE (b, c)`.
+2. **F1.** Up-front guard in `do_drop_column` after the index-covering
+   guard: column in any `schema.unique_constraints` entry → refuse with
+   the derived name + `alter table T drop constraint <name>`.
+3. **F2.** `error.generic.hint_no_table` catalog entry; in
+   `translate_generic`, pick it when `ctx.table` is `None`.
+
+Test-first for each (reproduce → fail → fix → pass), across the worker
+API (Tier-1/3) and the friendly-layer unit tests + insta snapshots.
+
+## Phase 5 — verification
+
+Full `cargo test` + clippy; compare to baseline; every checklist item
+addressed; engine-neutral vocab held (no SQLite/STRICT/PRAGMA in new
+user-facing strings); ADR + README + this plan lockstep.