I made a preliminary review of this PR. Take it with a grain of salt, I still need to run some tests. Thanks for this first attempt.

Reviewed 3 of 3 files at r1.
Reviewable status: complete! 0 of 0 LGTMs obtained

pkg/sql/sem/tree/datum.go, line 2495 at r1 (raw file):

	typ types.TTuple

	row bool

I believe you do not need this field any more.

pkg/sql/sem/tree/datum.go, line 2678 at r1 (raw file):

	if ctx.HasFlags(FmtParsable) {
		ctx.WriteString("ROW")
		if len(d.D) == 0 {

Empty tuples are so rare that you can drop the if len(d.D) == 0 case here. The formatting of the tuple below will produce the right output.

pkg/sql/sem/tree/eval.go, line 3787 at r1 (raw file):

		tuple.D[i] = d
	}
	tuple.row = t.Row

ditto uneeded

pkg/sql/sem/tree/expr.go, line 418 at r1 (raw file):

		binExprFmtWithParenAndSubOp(ctx, node.Left, node.SubOperator.String(), opStr, node.Right)
	} else {
		if t, ok := node.Right.(*Tuple); ok {

You need the special case for the suboperator case above too.

pkg/sql/sem/tree/expr.go, line 423 at r1 (raw file):

			ctx.WriteString(opStr)
			ctx.WriteByte(' ')
			ctx.WriteByte('(')

Combine the two WriteByte calls. Also, you need ROW here if and only if the oeprator is not IN/NOT IN

pkg/sql/sem/tree/expr.go, line 770 at r1 (raw file):

	}
	if ctx.HasFlags(FmtParsable) {
		ctx.WriteString("ROW")

That's too many occurrences of ROW, IMHO.