Performance: multi-threaded levelized timing propagation (std-only)

[shivarammysore]

Context
Arrival/required propagation is single-threaded. Large routed blocks leave cores idle.

Task
Levelize the timing graph (topological levels) and propagate independent nodes in parallel within each level; do the backward (required) pass the same way. Keep the engine dependency-free — use std::thread::scope, no external crates. Level barriers make results order-independent and deterministic.

API

• --threads N (default = available parallelism; --threads 1 = current behaviour).

Acceptance

• bit-identical results vs single-threaded across the whole test suite (determinism test);
• measured speed-up on a large routed block (report scaling);
• no new dependencies.