Fix inconsistent L* values in make_lstar1 caused by persistent Ilflag state

[lkncl]

fixes #69
fixes #41

Summary

This PR fixes a bug where calling make_lstar1 with identical inputs returns inconsistent L* values.
The issue originates from the COMMON /flag_L/ Ilflag variable, which is used inside calcul_Lstar_opt to decide whether to reuse a predicted foot-point (when Ilflag = 1) or compute a new linear estimate (when Ilflag = 0).

Previously:

Ilflag was only initialized to 0 inside make_lstar1.
After the first successful L* computation, Ilflag became 1 and was not reset before subsequent calls.
This caused calcul_Lstar_opt to reuse the last theta value from previous calls, leading to inconsistent L* results for identical inputs.
This bug only affects make_lstar1: all other routines calling calcul_Lstar_opt do not rely on the returned value of Ilflag.

grep -rnw source/ -e "calcul_Lstar_opt"
source/calcul_Lstar_o.f:37:       call calcul_Lstar_opt(t_resol,r_resol,
source/calcul_Lstar_o.f:42:       SUBROUTINE calcul_Lstar_opt(t_resol,r_resol,
source/AE8_AP8.f:163:      CALL calcul_Lstar_opt(t_resol,r_resol,xGEO
source/onera_desp_lib.f:130:           CALL calcul_Lstar_opt(t_resol,r_resol,XGeo
source/onera_desp_lib.f:238:           CALL calcul_Lstar_opt(t_resol,r_resol,xGEO
source/onera_desp_lib.f:246:               CALL calcul_Lstar_opt(t_resol,r_resol,xGEOp(1,ipa)
source/onera_desp_lib.f:1761:           CALL calcul_Lstar_opt(t_resol,r_resol,xGeo
source/onera_desp_lib.f:1774:               CALL calcul_Lstar_opt(t_resol,r_resol,xGEOp(1,ipa)
source/AFRL_CRRES_models.f:143:           CALL calcul_Lstar_opt(t_resol,r_resol,xGEO

I have checked that the logic around these calls does not involves Ilflag. Moreoever, make_lstar_shell_splitting1, make_lstar_shell_splitting2 already have a Ilflag=0 line before each call.

What this PR changes
✔️ 1. Removed the handling of Ilflag from the caller (make_lstar1)

make_lstar1 no longer uses Ilflag
The logic related to Ilflag_old and the fallback call has been removed.

✔️ 2. Ilflag is now always initialized to 0 inside calcul_Lstar_opt

This ensures deterministic behavior: each computation starts cleanly and does not reuse stale state.

✔️ 3. Removed the conditional branch based on Ilflag

Outcome

After this fix, calling make_lstar1 multiple times with identical parameters now returns consistent L*, Lm, B, Beq, I, and MLT values.