Align `cx_ecfp_add_point_no_throw` errors with OS

[bigspider]

The following code has different behaviours on speculos versus on a real device:

    // P = 0479be667ef9dcbbac55a06295ce870b07029bfcdb2dce28d959f2815b16f81798483ada7726a3c4655da4fbfc0e1108a8fd17b448a68554199c47d08ffb10d4b8
    uint8_t P[65] = {0x04, 0x79, 0xbe, 0x66, 0x7e, 0xf9, 0xdc, 0xbb, 0xac, 0x55, 0xa0, 0x62, 0x95, 0xce, 0x87, 0x0b, 0x07, 0x02, 0x9b, 0xfc, 0xdb, 0x2d, 0xce, 0x28, 0xd9, 0x59, 0xf2, 0x81, 0x5b, 0x16, 0xf8, 0x17, 0x98, 0x48, 0x3a, 0xda, 0x77, 0x26, 0xa3, 0xc4, 0x65, 0x5d, 0xa4, 0xfb, 0xfc, 0x0e, 0x11, 0x08, 0xa8, 0xfd, 0x17, 0xb4, 0x48, 0xa6, 0x85, 0x54, 0x19, 0x9c, 0x47, 0xd0, 0x8f, 0xfb, 0x10, 0xd4, 0xb8};
    // Q = 0479be667ef9dcbbac55a06295ce870b07029bfcdb2dce28d959f2815b16f81798b7c52588d95c3b9aa25b0403f1eef75702e84bb7597aabe663b82f6f04ef2777
    uint8_t Q[65] = {0x04, 0x79, 0xbe, 0x66, 0x7e, 0xf9, 0xdc, 0xbb, 0xac, 0x55, 0xa0, 0x62, 0x95, 0xce, 0x87, 0x0b, 0x07, 0x02, 0x9b, 0xfc, 0xdb, 0x2d, 0xce, 0x28, 0xd9, 0x59, 0xf2, 0x81, 0x5b, 0x16, 0xf8, 0x17, 0x98, 0xb7, 0xc5, 0x25, 0x88, 0xd9, 0x5c, 0x3b, 0x9a, 0xa2, 0x5b, 0x04, 0x03, 0xf1, 0xee, 0xf7, 0x57, 0x02, 0xe8, 0x4b, 0xb7, 0x59, 0x7a, 0xab, 0xe6, 0x63, 0xb8, 0x2f, 0x6f, 0x04, 0xef, 0x27, 0x77};
    uint8_t R[65];   

    uint32_t res = cx_ecfp_add_point_no_throw(CX_CURVE_SECP256K1, R, P, Q);
    PRINTF("res=%08x\n", res);

This returns CX_INTERNAL_ERROR = 0xFFFFFF85 on speculos, while it's CX_EC_INFINITE_POINT = 0xFFFFFF41 on a real device - which is the correct behaviour, as the two points are one the negation of the other so P + (-P) is the point at infinity.

Error situations also have mismatching results, but this one is particularly annoying as it's potentially a legitimate operation, rather than an 'error'.

[bigspider]

The issue is probably on this line. The point at infinity is not an error as per the SSL point addition, but conversion to affine coordinates is impossible and fails with the generic error. The point at infinity should maybe be tested before calling cx_ecpoint_from_EC_POINT, or (maybe cleaner), cx_ecpoint_from_EC_POINT could test for the point at infinity and return CX_EC_INFINITE_POINT as the "error" in that case.