AlgoScan

_vn_to_register, _const_vn, and _comparison_exprs were shared globally
across the dominator-tree walk. This meant a register-to-register copy
in one branch could be replaced with a representative register from a
sibling branch (via shared constant VNs), violating SSA dominance.

Bump pyjwt and requests to resolve pip-audit CVEs

Implementation of Crowdfund use case in Cairo

implementation of simple_transfer use case in leo

A warning is emitted if these are not explicitly declared when there is a map, but if that warning is ignored, we shouldn't count them towards the state totals. It would mean that inserting the first map value would succeed in testing, but not the second, and it would be better to fail fast.

A warning is emitted if these are not explicitly declared when there is a map, but if that warning is ignored, we shouldn't count them towards the state totals. It would mean that inserting the first map value would succeed in testing, but not the second, and it would be better to fail fast.

When a block has mutually-dependent phi nodes (e.g. from a swap or
shift-register pattern in a loop) and is reduced to a single
predecessor, collapsing those phis to assignments would produce a
use-before-def: a register appears as both target and source in the
same parallel copy, but the predecessor that provided the bootstrap
values is gone.

No current optimization can trigger this — these checks are
defense-in-depth so that future changes to the optimizer surface a
clear error rather than silently producing incorrect IR.