Major speedup to simulations

[dsharlet]

This change makes the solver much more aggressive about lifting solutions out of the non-linear solver. Previously, we would only search for solutions the order produced by RowReduce. Now, we instead repeatedly search for solutions that are independent of any remaining unknowns.

This change is a result of discussion with @andy-cytomic!

The impact to sim speed on example circuits is as follows:

Circuit	master	solver-fix
59 Bassman Preamp+Tone Stack	128.4	129.1	1.005451713
Big Muff Pi	77.49	90.2	1.164021164
Boss Super Overdrive SD-1	505	779.6	1.543762376
Bridge Rectifier	688.6	812.4	1.179785071
Common Cathode Triode Amplifier	328.6	438.8	1.335362142
Common Emitter Transistor Amplifier	921.2	1294	1.404689535
Dunlop Cry Baby GCB-95	180.6	308.1	1.705980066
Fender 5e3	33.67	41.96	1.246213246
Ibanez Tube Screamer TS-9	837.8	1285	1.533778945
Marshall Blues Breaker	351.6	521.3	1.482650739
Marshall JCM2000 DSL Preamp	47.72	59.91	1.255448449
Marshall JCM800 2203 preamp modded	46.42	59.13	1.273804395
Marshall JCM800 2203 Preamp	50.25	61.24	1.218706468
MXR Distortion +	438.7	522.8	1.191702758
MXR Phase 90	31.37	38.07	1.213579853
Op-Amp Model	681.4	1482	2.174933959
Orange Rockerverb 50 Preamp	44.01	52.51	1.193137923
Passive 1stOrder Lowpass RC	3.01E+04	3.04E+04	1.01E+00
Pro Co Rat	126.2	161.1	1.276545166
Wien Bridge Oscillator	645.3	1050	1.627150163

[dsharlet]

I think this is largely addressing #199