MPLipidAccessibility fails when the entire pose is outside the membrane

From discussion #163

core::membrane::convex_hull() causes a segfault in core::Size minx = outside[ tmp_x.index( numeric::min( tmp_x ) ) ]; because the tmp_x vector is empty. This is because none of the coordinates pass the coords.at( i ).z() >= min_z && coords.at( i ).z() <= max_z test previously, as the min_z and max_z settings are well outside the range of the protein coordinates (-15 to -5, where almost all the protein z-coordinates are positive).

Part of the reason may be a non-consistent span file. (But even then, the code shouldn't segfault.)

One question is if the reliance on absolute Cartesian z coordinates here is warranted, or if what we really want is the membrane-transformed z offset (is the assumption that the membrane is centered on z=0?) -- But even if that's corrected, the code should be robust to 'membrane' proteins which aren't actually in the membrane.

Reproduction steps:

• Download 2A65.pdb from the wwPDB
• Remove non-protein residues
• Use the following span file

#TM region prediction for BRD4 predicted using OCTOPUS
12 510
antiparallel
n2c
12      32
41      61
92      122
165     185
197     217
244     264
283     303
334     364
377     397
406     426
450     470
485     505

And options:

-in:membrane
-out:pdb
-mp:scoring:hbond
-mp:setup:spanfiles LeuT.span
-mp:setup:transform_into_membrane
-mp:transform:optimize_embedding
-score:weights=mpframework_smooth_fa_2012.wts

with command

score_jd2.linuxgccdebug -s 2A65.pdb @options

Reproduce the failure with 2A65.pdb, the supplied LeuT.span file, options, and score_jd2.linuxgccdebug. Start in core::membrane::convex_hull() at the empty tmp_x access, then inspect the preceding z-coordinate filtering; done means the out-of-membrane case no longer segfaults and the behavior is covered by an appropriate test.