feat: XFOIL objectives - minimum Cd at target Cl, maximum Cl with Cd limit

[PSamuele]

What

New prompt in the XFOIL solver:

1. Minimum Cd at a target Cl (asks the target Cl)
2. Maximum Cl with a Cd limit (asks the maximum Cd)

Score:

• Mode 1: Cd in drag counts (1 count = 0.0001). If Cl is outside target +/- 0.005, a quadratic penalty is added (0.001 of Cl outside the tolerance = 100 counts).
• Mode 2: -Cl. If Cd is above the limit, a quadratic penalty is added.

Other changes:

• The chosen objective is printed before the table. A warning is printed if the final Cl is outside the tolerance (mode 1) or the final Cd is above the limit (mode 2).
• Results folder: ..._Cl<target> (mode 1) or ..._CdMax<limit> (mode 2).
• Score column one character wider in both programs (negative scores in mode 2). The history plot uses a symmetric log scale when scores are negative.
• Failed analyses and airfoils outside the box now score 1e9 instead of 1e6. With the new penalties a valid airfoil far from the target Cl could score up to about 4e8, i.e. worse than a failed analysis. The Genetic Algorithm would then prefer failed analyses to valid airfoils.
• README updated.

The in-house solver is unchanged: it cannot minimise Cd until a boundary-layer model is added.

Results (Re 3.42e6, alpha 4 deg, seed 42)

	Cl	Cd
Previous PR (Cl target only), sandbox	0.8009	0.00656
Mode 1, target Cl 0.8, Windows	0.7967	0.00571
Mode 2, Cd <= 0.007, Windows (run with failure score 1e6)	1.5922	0.00688
Mode 2, Cd <= 0.007, sandbox (failure score 1e9)	1.5802	0.00699

Notes

• Mode 1: the best airfoil is found by the Genetic Algorithm; SLSQP rarely improves it. XFOIL prints Cd with 5 decimals (0.1 count), so the finite-difference gradient of Cd is coarse.
• Mode 2: with the default limit (Cd 0.02) the constraint is not active at 4 deg (final Cd 0.007). Maximum Cl is reached by pushing the camber to its upper bound (m = 0.09). A lower default limit may be more useful: to be decided.
• Open question from the previous PR: now that mode 1 has a well-defined objective, should the default seed be fixed instead of random?

Verified

• Baseline 24/24.
• Windows: both modes run; final Cl within tolerance (mode 1) and Cd below the limit (mode 2).
• Sandbox: with failure score 1e9, every valid airfoil scores below every failed one (max valid 3.7e7, min failed 1.0e9).

Depends on

#8 (base branch: feat/optimizer-parity).