Function evalTargetForSelection

What a text selection in a formula resolves to — the eval tip's target, and F9's.

Three rules, in order, because no single one is right:

1. Exactly a subtree → evaluate it in place (evalIndex). The common case: selecting SUM(A1:A5) inside =SUM(A1:A5)*2.

2. A whole-token run that parses on its own → evaluate that TEXT as its own expression. The subtree rule alone is not enough, because an operator chain is left-nested: =1+1+1+1 is (((1+1)+1)+1), so the middle 1+1 is not any node — its two operands live in different subtrees. Rule 1 would widen to 1+1+1 and answer 3 for a selection that plainly reads as 2. Excel's F9 works this way throughout (it evaluates the selected text, not a node), and this is where that shows.

   Whole-token alignment is what keeps this honest: 1*2 inside =A1*2 parses, but it starts mid-token (A1), so it never reaches this rule and can't invent a value out of a coincidence.

3. Otherwise → snap OUTWARD to the smallest evaluable segment containing the selection. A drag that lands mid-token (UM(A1:A5 in =SUM(A1:A5)*2) still answers — for SUM(A1:A5) — and the host's outline shows that it widened. Requiring a clean selection instead would leave the tip silent most of the time a human made the selection by hand.

null when the selection touches nothing evaluable (whitespace only, an unparsed tail, or a selection past the tokens).