fix(parser): remove duplicate [T] array-sugar production (#214)

`type_expr_primary` declared the array-shorthand production
`LBRACKET type_expr RBRACKET -> TyApp("Array", ...)` twice (lines
443/469) — added independently by issues-drafts/02 and #40 with
byte-identical semantic actions. The second copy was a verbatim
duplicate of the first.

Effects of the duplication, both confirmed via `menhir --explain`:

  - Menhir reported "1 production is never reduced" (the second copy
    is dead — the first always wins in the LALR automaton).
  - It generated a real reduce/reduce conflict family: state 80
    (reached after `[ type_expr ]`) could not choose which identical
    production to reduce, and that ambiguity propagated through
    inherited lookaheads.

Removing the dead copy and folding its substantive rationale (the
`Array`-not-`List` desugar reason + typecheck.ml canonicalisation
note + #40 ref) into the surviving comment so no documented knowledge
is lost.

Measured impact (full `dune test --force` gate green at 257/257 before
and after — zero parse-behaviour change, this is a pure dedup):

  - "production never reduced" warning: eliminated.
  - reduce/reduce conflicts: 36 -> 10 (R/R conflict states 5 -> 4).
  - shift/reduce conflicts: unchanged at 75 (the duplicate was a
    reduce/reduce source only).

The residual 75 S/R + 10 R/R conflicts are inherent LALR(1)
ambiguities that Menhir resolves deterministically and correctly (the
257-test suite, incl. STAGE-A AOT + multi-module + STAGE-B effect-row
coverage, proves the chosen resolution is the intended language).
This Menhir (20260209) has no expected-conflict declaration mechanism,
so eliminating the rest requires deliberate per-conflict grammar
refactoring with estate-wide parse blast radius — tracked separately,
not bundled here.

Refs #40

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

Author: hyperpolymath

lib/parser.mly

@@ -437,9 +437,14 @@ type_expr_primary:
      LT/GT tokens are shared with expression-position less-than. */
   | name = upper_ident LT args = separated_nonempty_list(COMMA, type_arg) GT
     { TyApp (mk_ident name $startpos(name) $endpos(name), args) }
-  /* Array sugar: [T] desugars to Array[T] (issues-drafts/02). The element
-     type can be any type_expr (recursive), so [[Int]] means Array[Array[Int]]
-     and [Result[T, E]] works as expected. */
+  /* Array sugar: `[T]` desugars to `Array[T]` (issues-drafts/02; #40). The
+     element type can be any type_expr (recursive), so `[[Int]]` means
+     Array[Array[Int]] and `[Result[T, E]]` works as expected. This is the
+     syntax stdlib has used all along (`fn map<T, U>(arr: [T], ...)`). The
+     typechecker (lib/typecheck.ml ~724, 813, 1024) canonicalises array
+     literals/operations on `TApp (TCon "Array", ...)`, so `Array` is the
+     right desugar target — `List` here would trigger a
+     `Unify.TypeMismatch (List, Array)` at check time. */
   | LBRACKET elem = type_expr RBRACKET
     { TyApp (mk_ident "Array" $startpos $endpos, [TyArg elem]) }
   /* Function-type-as-type: `fn(A, B) -> C` lowers to the curried arrow
@@ -458,16 +463,6 @@ type_expr_primary:
      the interior in one pass without lookahead conflicts. */
   | LBRACE body = ty_record_body RBRACE
     { TyRecord (fst body, snd body) }
-  /* Array shorthand: `[T]` desugars to `Array[T]`.  This is the syntax
-     stdlib has used all along (`fn map<T, U>(arr: [T], f: T -> U) -> [U]`,
-     `Result[[T], E]`, etc.) but it was previously only accepted in stdlib
-     load paths, not in user source.  The typechecker (lib/typecheck.ml
-     lines 724, 813, 1024) canonicalises array literals/operations on
-     `TApp (TCon "Array", ...)`, so `Array` is the right desugar target —
-     using `List` here triggers a `Unify.TypeMismatch (List, Array)` at
-     check time. Closes #40. */
-  | LBRACKET ty = type_expr RBRACKET
-    { TyApp (mk_ident "Array" $startpos $endpos, [TyArg ty]) }
   /* Built-in types */
   | NAT { TyCon (mk_ident "Nat" $startpos $endpos) }
   | INT_T { TyCon (mk_ident "Int" $startpos $endpos) }