perf: vectorized panel method (same results, about 50x faster)

[PSamuele]

What

_calculate_influence_coefficients in naca_core/panel_method.py computed the influence coefficients with a Python double loop (160 x 160 = 25,600 iterations per analysis).
The same formulas are now evaluated for all panel pairs at once with NumPy arrays. The diagonal (i == j) is zero, as before. No other function is changed.

Verified

• Baseline 24/24 on Windows and Linux, max difference 2e-15 (round-off only).
• Linux: baseline also run with warnings as errors (-W error): no division-by-zero warnings.
• Windows: in-house optimizer with seed 136031 gives exactly the same run as before (112 iterations, NACA 3207, m = 0.0308, p = 0.2392, t = 0.0687).
• Linux: in-house optimizer with seed 42, all 105 table rows identical to the old version.

Speed (Linux sandbox)

	Before	After
One analysis, 160 panels	0.247 s	0.0046 s
Full in-house optimization	30.3 s	2.9 s

Measured on a single-core Intel Xeon 2.10 GHz VM, Ubuntu 24.04, Python 3.12.3, numpy 2.4.4 (OpenBLAS). Absolute times depend on the machine; the ratio does not.

The XFOIL optimizer is almost unaffected: its time is spent in XFOIL.

Depends on

#10 (base branch: feat/validation-error-breakdown).

[Filettus-newkeyboard]

Mi sono preso del tempo per chiedere a Gemini cosa cambia, e adesso ho imparato come il comando python np.asarray vettorializza e migliora tutto... Niente da dire... Chapeau Samu!!!

[PSamuele]

Mi sono preso del tempo per chiedere a Gemini cosa cambia, e adesso ho imparato come il comando python np.asarray vettorializza e migliora tutto... Niente da dire... Chapeau Samu!!!

Calma prima di fare complimenti, claude mi ha spiegato di np.asarray, io ho chiesto quali fossero i limiti strutturali del codice riguardo alla velocità e come migliorarla...