Length of an array of Variant measures its first element

Introduced by c65fc287f6 (fix(A): Length, UpCase and Copy of a Variant work on the string it holds). Bisected: c65fc287f6^ prints 35 7 9 and c65fc287f6 prints 30 1 9 on test_open_array_of_variant.

The Variant arm in the Length argument lowering (ir.inc) asked ASTTk[arg] = tyVariant. Arrays carry their element kind, so it could not tell Length(v) from Length(arrOfVariant). The same lesson is recorded on ASTNodeIsWholeArray, for interfaces. The arm now asks that predicate and NodeDynDepth as well.

test_length_of_an_array_of_variant covers a dynamic, open (from dyn, static and bracket arguments), field and static array; Copy of an array of Variant; and the Variant string ops the original commit added. Its expected output is fpc 3.2.2's, built with cwstring. On c65fc287f6, five of its eight lines are wrong. It passes on six targets. The fixture from the original commit, test_length_of_a_variant, still passes, and so does the NilPy row.

The quick gate runs neither borg row, which is how this got past it.