Add opt-in --score_coefficients and --write_score_descriptors

[ksy141]

Adds two opt-in options to fpocket. Without them, behaviour and output are unchanged.

--score_coefficients c0,...,c8 replaces the pocket score by a linear score over nine terms:

c0 + c1*nas_norm + c2*prop_asapol_norm + c3*mean_loc_hyd_dens_norm + c4*polarity_score
   + c5*as_density + c6*convex_hull_volume + c7*surf_pol_vdw14 + c8*surf_apol_vdw14

fpocket 4's built-in score (the last assignment in score_pocket) has exactly this form. Passing its coefficients, -0.03783394,0.48461469,0,0,0,0.09093926,0.0004155899,-0.003995233,-0.004072336, reproduces the default scores and ranking exactly. The option exists so the ranking can be refitted for structures that differ from the all-atom training data, without recompiling. We use it for coarse-grained proteins (Martini 2, Martini 3, SIRAH), where the default score ranks pockets poorly. We refitted the score on fpocket's own training set (fpocket-data-1.0, 263 complexes) and retuned the detection flags. Together they raise Top-1 on fpocket's 48 apo test structures from about 0.1 to about 0.6.

--write_score_descriptors writes those eight descriptors for every pocket to <name>_out/<name>_score_desc.txt, so a score can be refitted from one fpocket run per structure.

Changes:

• fparams.h / fparams.c: the two long options, parsing (exits with a message unless exactly 9 numbers are given), and -h text.
• pscoring.c: score_pocket_custom().
• pocket.c: use it when the option is set.
• fpout.c / fpmain.c: the descriptor file.

Testing:

• Default output is identical to the current master on 39 structures, apart from the Monte-Carlo volume lines, which vary between runs anyway.
• With the built-in coefficients passed explicitly, scores are identical to the default (for example 22/22 pockets on 11GS).
• The repository's tests give the same results as on master; 8 of 11 fail there too, before this change.