fix(lexer): {array}[index] no longer mis-tokenises [ as bracket-string

The lexer's isStringBracket disambiguator decides whether `[` opens
an indexing operator or a Harbour bracket-string literal. The
heuristic checks the previous token's kind and treats the bracket
as indexing only when preceded by an IDENT, RPAREN, RBRACKET, or a
literal. RBRACE was missing — so

   FieldPut(3, {"Kim","Lee","Park","Choi","Yoon"}[Int(Mod(i-1,5))+1])

tokenised the `[` after `}` as a bracket-string opener, swallowed
through the first `]` it found, and produced bogus parse errors
("expected ), got STRING …"). RBRACE is now in the indexing-context
set, so an inline array-literal followed by `[index]` works.

Surfaced by the examples/ build sweep — fixed test_all_rdd,
test_index_adv, test_multi_rdd, test_rdd_full all in one go.

The sweep itself is committed as tests/examples_build.sh — builds
every PRG under examples/ and reports any compiler / preprocessor
errors. Run it after compiler changes to catch regressions in
broad-coverage user-style code that the focused suites don't
exercise.

Current sweep state: 65 / 71 examples build cleanly. The remaining
6 failures are all #pragma BEGINDUMP blocks that import external
Go packages (http, websocket, sqlite, time) — not Five-side bugs.

Other gates green:
  go test ./...      : PASS
  FiveSql2 SQL:1999  : 43/43
  Harbour compat     : 56/56
  std.ch suite       : 16/16
  FRB suite          : 7/7
  examples build     : 65/71 (rest = external Go deps)

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

compiler/lexer/lexer.go

@@ -330,10 +330,11 @@ func (l *Lexer) scanString(quote byte) token.Token {
 }
 
 // isStringBracket returns true if [ should be treated as string delimiter.
-// Harbour: [text] is string when not preceded by ident, ), ], literal.
+// Harbour: [text] is string when not preceded by ident, ), ], literal,
+// or `}` (close of array literal — `{1,2,3}[2]` is index access).
 func (l *Lexer) isStringBracket() bool {
 	switch l.lastKind {
-	case token.IDENT, token.RPAREN, token.RBRACKET,
+	case token.IDENT, token.RPAREN, token.RBRACKET, token.RBRACE,
 		token.INT, token.LONG, token.DOUBLE, token.STRING,
 		token.TRUE, token.FALSE, token.NIL_LIT:
 		return false // array index context

tests/examples_build.sh

@@ -0,0 +1,42 @@
+#!/usr/bin/env bash
+# Build-only sweep of every example PRG. Doesn't run them (many are
+# interactive / database / network). Just checks the compiler accepts
+# them cleanly.
+set -e
+cd /Users/charleskwon/Projects/fivedev/five
+
+work=$(mktemp -d)
+trap 'rm -rf "$work"' EXIT
+
+pass=0
+fail=0
+declare -a failed
+
+for src in examples/*.prg; do
+   name=$(basename "$src" .prg)
+   if ./five build "$src" -o "$work/$name" >"$work/$name.out" 2>&1; then
+      # Check for parse / preprocessor errors even when build "succeeded"
+      if grep -qE '^(five: [0-9]+ (parse|preprocessor) error|panic:)' "$work/$name.out"; then
+         echo "FAIL $name (build returned 0 but errors detected)"
+         tail -5 "$work/$name.out" | sed 's/^/   /'
+         fail=$((fail+1))
+         failed+=("$name")
+      else
+         pass=$((pass+1))
+      fi
+   else
+      echo "FAIL $name"
+      tail -5 "$work/$name.out" | sed 's/^/   /'
+      fail=$((fail+1))
+      failed+=("$name")
+   fi
+done
+
+echo
+echo "================================================================"
+echo "  Build sweep: $pass / $((pass+fail)) examples build cleanly"
+echo "================================================================"
+if [ ${#failed[@]} -gt 0 ]; then
+   echo "Failed:"
+   for n in "${failed[@]}"; do echo "  $n"; done
+fi