fix(parser): resolve state-41 effect-row MINUS/ARROW S/R conflict via precedence

First PR of the dedicated grammar-conflict-elimination workstream
(epic #215), family A (effect-row MINUS overload), owner-prioritised
first.

In `type_expr_arrow`, after a `type_expr_primary`, lookahead `ARROW`
(continue an arrow type) or `MINUS` (open the legacy `-{E}->` effect
row, parser.mly:410) raced against reducing the base case
`type_expr_arrow -> type_expr_primary` (state 41, tokens MINUS ARROW).
Shifting is *always* the correct choice (a bare type can never both be
followed by `->`/`-{` and want to stop there), and was already Menhir's
arbitrary resolution — so making it explicit changes no parse.

Fix: a lowest-precedence pseudo-token `LOWEST_TYPE_ARROW` (never lexed)
tags the base reduction; `ARROW` is given a precedence just above it.
`MINUS` already outranks both, so on either lookahead the parser shifts
*by declared precedence* instead of by an arbitrary tie-break, and the
conflict disappears. `ARROW` is a conflict lookahead ONLY in state 41
(verified via `menhir --explain`), so giving it precedence is
side-effect-free.

Rigorous verification (full `menhir --explain` diff, baseline = #214):

  - S/R arbitrarily resolved: 75 -> 72
  - S/R conflict states:      25 -> 23 (state 41 eliminated)
  - R/R: 10 (untouched); no new conflict states introduced
  - The 16 pre-existing "precedence/`%prec` never useful" warnings are
    byte-for-byte identical before and after — NOT introduced here
    (they are dead precedence declarations; filed as family F on #215).
  - `dune test --force` green at 257/257 (parse-behaviour-preserving).

Refs #215

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>

Author: hyperpolymath

lib/parser.mly

@@ -83,6 +83,20 @@ let rec effect_union_of_list = function
 %token EOF
 
 /* Precedence (lowest to highest) */
+/* `LOWEST_TYPE_ARROW` is a pseudo-token (never lexed) carrying the
+   lowest precedence. It tags the base `type_expr_arrow -> type_expr_primary`
+   reduction so that in state 41 — after a `type_expr_primary`, lookahead
+   `ARROW` (continue an arrow type) or `MINUS` (open the legacy `-{E}->`
+   effect row) — the parser SHIFTS rather than reducing the base case.
+   Shift was already Menhir's (correct) arbitrary resolution there (a
+   bare type can never be followed by `->`/`-{` *and* want to stop), so
+   declaring this precedence is parse-behaviour-preserving and only
+   silences the spurious conflict. `ARROW` has no precedence elsewhere
+   and is a conflict lookahead ONLY in state 41 (verified via
+   `menhir --explain`), so giving it precedence here is side-effect-free.
+   See affinescript#215 (family A). */
+%nonassoc LOWEST_TYPE_ARROW
+%nonassoc ARROW
 %right EQ PLUSEQ MINUSEQ STAREQ SLASHEQ
 %left PIPEPIPE
 %left AMPAMP
@@ -416,7 +430,7 @@ type_expr_arrow:
      past the closing RPAREN. */
   | LPAREN ty1 = type_expr COMMA tys = separated_nonempty_list(COMMA, type_expr) RPAREN ARROW ret = type_expr_arrow
     { List.fold_right (fun p acc -> TyArrow (p, None, acc, None)) (ty1 :: tys) ret }
-  | ty = type_expr_primary { ty }
+  | ty = type_expr_primary %prec LOWEST_TYPE_ARROW { ty }
 
 type_expr_primary:
   | LPAREN RPAREN { TyTuple [] }