fix(pp,parser,gengo): pre-release blocker round (Wave 1)

Six audit-driven blockers landed together because they're tangled:

  * MENU TO removed from std.ch — the rule expanded to a call to a
    nonexistent __MenuTo() RTL symbol, so any user code with `MENU
    TO choice` compiled clean and panicked at runtime. Behavior
    pre-this-round was a parser silent no-op, which is at least
    consistent. Restore that until @ PROMPT (the companion command)
    actually lands.

  * COUNT now requires `TO <var>`. The earlier `[TO <v>]` optional
    bracket was a Harbour-pattern transcription error: the result
    template references `<v>` unconditionally, so a bare `COUNT`
    expanded to ungrammatical ` := 0 ; dbEval(...)` and the
    PRG parser rejected it. Match Harbour's std.ch which makes TO
    mandatory.

  * UPDATE FROM ... REPLACE now requires `FROM`/`ON`/`REPLACE` all
    three. Same root cause as COUNT: the result template uses
    `<key>`, `<f1>`, `<x1>` unconditionally; missing any of them
    produced broken syntax. Tightened to fail loudly rather than
    silently mis-expand.

  * CLOSE <unknown_alias> no longer closes the *current* workarea.
    SelectByAlias was a silent no-op when the alias was missing,
    leaving WASaveAndSelectAlias to evaluate the inner DbCloseArea()
    against the originally-selected WA — a real data-loss footgun.
    SelectByAlias now returns bool; WASaveAndSelectAlias switches to
    the no-area sentinel (0) on miss so the inner expression's
    Current() returns nil and short-circuits.

  * SUM <x1>, <xN> TO <v1>, <vN> — multi-pair form supported.
    Required two pieces:

       1. matchSegment's regular-marker stop-boundary now combines
          outerTail literals AND the segment's repeat boundary so
          `[, <xN>]` doesn't let `<xN>` swallow past the next ','.

       2. **Five parser miscompiled comma-separated expressions in
          code blocks.** `{|| e1, e2, e3 }` kept only the last expr
          and threw away earlier ones at *AST level*, so all their
          side effects vanished. New SeqExpr AST node + emitter
          (emit each, pop intermediate results) + folding/walk
          updates fix the underlying bug, which also unbreaks any
          other block that relied on comma sequencing.

  * pp.go's `;` continuation joiner now strips exactly one trailing
    `;` per iteration, preserving Harbour's `;;` convention (literal
    `;` followed by a continuation marker). Without this the SUM
    rule's chained `<v1> :=[ <vN> :=] 0 ; ; dbEval(...)` collapsed
    to a missing statement separator.

  * parseExprStmt's xBase fallback switch is back in sync with
    parseIdentStmt — COPY/SORT/COUNT/SUM/AVERAGE/TOTAL/UPDATE/JOIN/
    DISPLAY/LIST removed (std.ch handles all of them now). Leaving
    them in the fallback masked typos as silent no-ops.

Gates green:
  go test ./...      : PASS
  FiveSql2 SQL:1999  : 43/43
  Harbour compat     : 56/56

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

compiler/parser/expr.go

@@ -426,13 +426,24 @@ func (p *Parser) parseArrayOrBlock() ast.Expr {
 		}
 
 		// Parse block body — may have comma-separated expressions
-		// {|x| expr1, expr2} → comma = sequence, returns last value
-		body := p.parseExpr()
+		// {|x| expr1, expr2, expr3} → all evaluated in order, last is
+		// the return value. Earlier impl dropped intermediate exprs by
+		// overwriting `body`, which was a silent miscompile (any
+		// non-trailing side effect — e.g. `<v> := <v> + <x>` in a
+		// multi-pair SUM block — vanished).
+		first := p.parseExpr()
+		var seq []ast.Expr
 		for p.match(token.COMMA) {
-			// Comma-separated: wrap as sequence, keep last
-			body = p.parseExpr()
+			if seq == nil {
+				seq = []ast.Expr{first}
+			}
+			seq = append(seq, p.parseExpr())
 		}
 		rbrace := p.expect(token.RBRACE).Pos
+		var body ast.Expr = first
+		if seq != nil {
+			body = &ast.SeqExpr{Items: seq, StartAt: first.Pos(), EndAt: rbrace}
+		}
 
 		return &ast.BlockExpr{LBrace: lbrace, Params: params, Body: body, RBrace: rbrace}
 	}