Move the _is_provably_positive helpers out of tensor/subtensor.py #2404

_is_provably_positive and _is_provably_non_negative are general graph predicates that just happen to live in pytensor/tensor/subtensor.py. The module does use them itself for slice canonicalization, but five others import them from there — rewriting/subtensor.py, rewriting/math.py, rewriting/linalg/summary.py, assumptions/elemwise.py, and assumptions/positive_definite.py — so math rewrites and the assumptions system both pull in the indexing module just to ask whether a variable is non-negative.

No obvious destination. assumptions/ already holds two of the callers, but importing it from pytensor/tensor inverts the current direction.

Follow-up from #2380.