Use libRoadRunner saveState

[emitra17]

A libroadrunner feature is under development (see sys-bio/roadrunner#550) that allows saving and reloading a libroadrunner object. This would let us avoid the slow XML load that currently occurs for every simulation. I've tested a pre-release libroadrunner with this feature, and it is working well.

When this feature is released, PyBNF should use this workflow to gain a significant speedup in SBML fitting.

[wshlavacek]

Implemented: RoadRunner saveState/loadState optimization

PyBNF now caches the compiled RoadRunner state to a temp file during model initialization (__init__), then uses loadState instead of re-parsing SBML XML on every execute() and model_text() call.

Benchmark results (50 iterations each)

Model	XML load	State load	Speedup
parabola.xml	0.536s	0.331s	1.6x
Romano2014.xml	2.318s	0.773s	3.0x
yeast.xml	6.205s	1.517s	4.1x

Speedup scales with model complexity — larger SBML models benefit most.

Changes

• SbmlModelNoTimeout.__init__ now calls runner.saveState() to a temp file after the initial XML load
• New _load_runner() method creates a fresh RoadRunner() and calls loadState() — used by execute() and model_text()
• Minimum libroadrunner version bumped from 1.5.2 → 1.6.0 (the version that introduced saveState/loadState, per sys-bio/roadrunner#550)

Files modified: pybnf/pset.py, setup.py