Restore rigorous baseline validation with 20-seed statistical significance

[TasfinMahmud]

@abhiprd2000 The evaluation has been re-run with the seed count increased from 5 to 20 to address the low statistical power and establish formal significance. As anticipated, the standard error bounds tightened considerably with the larger sample size.

Here are the aggregated results from the 20-seed run:

============================================================
FINAL AGGREGATED RESULTS (20 Seeds)
============================================================
Physics GAP: +0.538 ± 0.136
Softmax GAP: +0.247 ± 0.215
MC-Drop GAP: +0.334 ± 0.091
Ensemble GAP: +0.386 ± 0.120

Statistical Significance (Physics vs Ensemble):
  Paired t-test p-value: 0.0032
  Average Matched Coverage: N = 5567.1 / 14914 (37.33%)

Noise Test Catch Rate (Physics catches Ensemble's confident errors):
   clean: 0.717 ± 0.273
    20dB: 0.726 ± 0.268
    10dB: 0.772 ± 0.237
     5dB: 0.829 ± 0.177
     0dB: 0.827 ± 0.176

With $N=20$, the paired t-test comparing the Physics GAP against the Deep Ensemble yields a p-value = 0.0032, successfully meeting the $p < 0.05$ threshold for statistical significance. The methodology is now rigorous and the script has been permanently updated to reflect the 20-seed requirement.