Multi-corner timing analysis (min + max SDF)

[robtaylor]

Summary

Run GPU simulation at both min and max SDF corners to catch both setup violations (slow/max corner) and hold violations (fast/min corner).

Current State

GEM picks a single SDF corner via --sdf-corner min|typ|max. This misses violations that only appear at the opposite corner.

Proposed Approach

• Run simulation twice with different corners, or support --sdf-corner all to load both and check against worst-case
• Report per-corner WNS/TNS

Impact

High — hold violations are invisible at typ/max corner. This is the most impactful timing improvement.

Effort

Low — can be done by running twice in CI initially, then optimized to single-pass later.