"BC" versions of TriSolve and TriMul.
PentaDiag still lacks this feature but for no-slip or free-slip, non-elliptical BC,
P=0 at the bondaries for the moment.

 + added time_SHT.c to measure speed and errors of SHT.c

SHTfast.c : developement continues...

timings for LMAX=170, MMAX=85, NLAT=256/512, NPHI=256 (ie 2/3 rule)
** on machine "calcul1", (SHT + iSHT) *1000, MRES=1
SHT, no polar opt   : 142
SHT, 1e-6 polor opt : 136 (-4%)
SHTfast (DCT)       : 106 (-25%)
** on machine "calcul1", (2 iSHT + NL(product) + 1 SHT), MRES=1
SHT, no polar opt   : 2096
SHT, 1e-6 polor opt : 1976 (-6%)
SHTfast (DCT)       : 1785 (-15%)
** on machine "calcul1", (2 iSHT + NL(product) + 1 SHT), MRES=2
SHT, no polar opt   : 1615
SHT, 1e-6 polor opt : 1467 (-10%)
SHTfast (DCT)       : 1560 (-3%) [bug: should be retimed]

=> spectral space NL term has poor performance (useful for MMAX < 8)

Added HS_to_grad_spat; 0:ilat/2 is now theta=0:pi/2; theta and phi derivatives tested OK (synthesis)

 - stability checked. (3000 transforms : error < 1e-10)
 - even/odd optimization leads to much improved performance !