Use previous slice as MG initial guess for explicit solver

[MaxThevenet]

This PR proposes to use Bx and By from the previous slice as an initial guess for the MG solver. We thought is was already the case, but apparently not. Here's a scan to see the improvement. As always, PC means predictor-corrector (old version), ii is now development and 00 indicates this PR. The improvement is marginal for this example.

output_0128PC000000.txt:TinyProfiler total time across processes [min...avg...max]: 1.639 ... 1.639 ... 1.639
output_0128WPmt0300.txt:TinyProfiler total time across processes [min...avg...max]: 1.496 ... 1.496 ... 1.496
output_0128WPmt03ii.txt:TinyProfiler total time across processes [min...avg...max]: 1.478 ... 1.478 ... 1.478
output_0128WPmt0400.txt:TinyProfiler total time across processes [min...avg...max]: 1.617 ... 1.617 ... 1.617
output_0128WPmt04ii.txt:TinyProfiler total time across processes [min...avg...max]: 1.84 ... 1.84 ... 1.84
output_0256PC000000.txt:TinyProfiler total time across processes [min...avg...max]: 4.398 ... 4.398 ... 4.398
output_0256WPmt0300.txt:TinyProfiler total time across processes [min...avg...max]: 3.105 ... 3.105 ... 3.105
output_0256WPmt03ii.txt:TinyProfiler total time across processes [min...avg...max]: 3.049 ... 3.049 ... 3.049
output_0256WPmt0400.txt:TinyProfiler total time across processes [min...avg...max]: 3.858 ... 3.858 ... 3.858
output_0256WPmt04ii.txt:TinyProfiler total time across processes [min...avg...max]: 4.008 ... 4.008 ... 4.008
output_0512PC000000.txt:TinyProfiler total time across processes [min...avg...max]: 19.2 ... 19.2 ... 19.2
output_0512WPmt0300.txt:TinyProfiler total time across processes [min...avg...max]: 7.91 ... 7.91 ... 7.91
output_0512WPmt03ii.txt:TinyProfiler total time across processes [min...avg...max]: 8.07 ... 8.07 ... 8.07
output_0512WPmt0400.txt:TinyProfiler total time across processes [min...avg...max]: 10.02 ... 10.02 ... 10.02
output_0512WPmt04ii.txt:TinyProfiler total time across processes [min...avg...max]: 10.26 ... 10.26 ... 10.26
output_1024PC000000.txt:TinyProfiler total time across processes [min...avg...max]: 143.8 ... 143.8 ... 143.8
output_1024WPmt0300.txt:TinyProfiler total time across processes [min...avg...max]: 32.39 ... 32.39 ... 32.39
output_1024WPmt03ii.txt:TinyProfiler total time across processes [min...avg...max]: 32.22 ... 32.22 ... 32.22
output_1024WPmt0400.txt:TinyProfiler total time across processes [min...avg...max]: 37.83 ... 37.83 ... 37.83
output_1024WPmt04ii.txt:TinyProfiler total time across processes [min...avg...max]: 37.89 ... 37.89 ... 37.89
output_2048PC000000.txt:TinyProfiler total time across processes [min...avg...max]: 2336 ... 2336 ... 2336
output_2048WPmt0300.txt:TinyProfiler total time across processes [min...avg...max]: 285.5 ... 285.5 ... 285.5
output_2048WPmt03ii.txt:TinyProfiler total time across processes [min...avg...max]: 287.1 ... 287.1 ... 287.1

Looking into output_0512WPmt04**.txt I get the following number of MG iterations

# iterations	old	new
1	0	0
2	10	41
3	501	470
4	0	0

So the new version is converging a bit more often with 2 iterations instead of 3, but it doesn't have a large impact on the runtime. Still, I think it is worth merging this, it could have a bigger impact in some cases (in particular when a lower tolerance is required).

[SeverinDiederichs]

Thanks for this PR!
Grabbing the last low hanging fruit of the MG solver