use the solver from each worker instead of a global one for model generation

[zapashcanon]

No description provided.

[zapashcanon]

I ran Test-Comp with -w24 and O3 and compared to our previous best run. tool1 is the previous version and tool2 is this PR.

tool1 had 593 reached and tool2 had 592 reached
tool1 had 535 timeout and tool2 had 530 timeout
tool1 had 075 nothing and tool2 had 075 nothing
tool1 had 001 other   and tool2 had 001 other
tool1 had 012 killed  and tool2 had 018 killed

tools have 572 reached tasks in common

tool1 had 021 tasks tool2 did not found
tool2 had 020 tasks tool1 did not found

on      commonly  reached tasks, tool 1 took 1591.708683 sec. (mean 2.782707) and tool 2 took 1312.187691 sec. (mean 2.294034)
on *not commonly* reached tasks, tool 1 took 169.633282 sec. (mean 8.077775) and tool 2 took 167.097363 sec. (mean 8.354868)

among tasks reached only by tool 1, tool 2 replied 000 nothing, 004 timeout, 00 other and 17 killed
among tasks reached only by tool 2, tool 1 replied 000 nothing, 009 timeout, 00 other and 11 killed

The results are quite similar but we are much faster on average, plus this allows to get rid of the global solver, which makes using other solvers easier. Thus I'm merging. :)