Add leaderboard entry: 0.0000 overlap, 0.3582 wirelength, 26.15s

[shaanpatel00]

Results (tests 1-10)

• Average Overlap: 0.0000
• Average Wirelength: 0.3582
• Total Runtime: 26.15s

Code: https://github.com/shaanpatel00/partcl_intern_challenge (public)
All changes are in placement.py between the OPTIMIZATION CODE markers.
test.py is unmodified.

Approach

1. IRLS analytic solve — the wirelength objective is convex and separable, so the unconstrained optimum is found via iteratively reweighted least squares (~3ms) rather than gradient descent (~1s for the same answer).
2. Pin-density-aware construction — standard cells hold most pins but almost no area (Test 10: 6.7% of area, 89% of pins), so they're packed into one tight cluster, with macros flush-packed around it at positions that minimize their own incident wirelength.
3. Randomized restart search over macro placement order (trial-count bounded, so results are reproducible).
4. Gradient refinement, kept only when it beats the construction and stays legal, then a legalization pass guaranteeing exact zero overlap.

Bonus tests

Test 11: 0.3135; Test 12: 0.5055, both zero overlap.
Note: calculate_cells_with_overlaps in the evaluation section is an O(N²) Python loop, so scoring test 12 takes ~45 min even though placement finishes in ~4s. I verified both with an independent bucket-based overlap checker.