[FTD] Simplifying FTD to One Suppression per Data Delivery

[QinYuan2000]

Motivation

The previous FTD implementation handled loops by splitting them into four distinct scenarios.
When a single data delivery simultaneously matched multiple scenarios, or when deliveries appeared inside nested loops, the pass inserted multiple cond_br to realize suppression.
This increased circuit area and control complexity, making behavior hard to reason about.

This PR focuses purely on simplification: each data delivery is guarded by exactly one suppression branch.
We adopt a unified handling that does not require classifying loop scenarios into four cases and does not stack multiple branches for overlapping situations.

Implementation

• Added local control dependence extraction between producer and consumer blocks, via a region-level subgraph with a virtual sink node.
  If the producer is traversed again, it is treated as the sink.
• In this subgraph, any loop scenario is flattened into a direct suppression path without nested loop handling.
  All such cases are managed through one unified suppression path.
• For any producer → consumer delivery, compute a single Boolean expression for suppression and insert one cond_br at the delivery site.
• For MUX conditions (see FTD paper), only consider MUX conditions that appear after the producer.