This removes an inconsistency between Numpy and Aesara broadcasting rules, where a variable dimension with unknown shape was always assumed to be non-broadcastable (i.e., different than 1)

This adds `SharedVariable` construction tracking that allows one to determine
which variables were created within a specific scope (e.g. within a Python
function).  With this ability, we're able to determine which shared variable
update should and shouldn't be performed within the iterations of a `Scan` node.

Co-authored-by: Brandon T. Willard <971601+brandonwillard@users.noreply.github.com>