Filter the artifact run's result file - #359

`simulate(variableFilter=…, resimulateExecutable=…)` keeps only
`simflags`: the experiment arguments reach a model through the build,
and a resimulate does not build. For the C lanes that is invisible,
since the executable reads the filter from the `_init.xml` its own
`buildModel` wrote, but a wasm artifact comes from `buildModelFMU`,
which has no `variableFilter` argument. Every wasm-jit runner has
therefore been writing every variable the model has. `artifactCmd`
already passes the rest of the experiment as simflags for this same
reason; `-variableFilter` was the one it missed.

Measured on the MSL 4.1.0 QuasiStatic FundamentalWave `IMC_Conveyor`,
whose comparisonSignals.txt names three variables:

|                                 | no filter     | -variableFilter |
| ------------------------------- | ------------- | --------------- |
| result file                     | 1162 MB, 2857 | 4.97 MB, 3      |
| timeSimulation                  | 8.81 s        | 7.50 s          |
| verification, 500-row reference | 1.2-2.0 s     | 0.42 s          |
| verification, peak RSS          | 2.34 GB       | 71 MB           |

Over the 10801 models both lanes verified in the latest run, master's
verification column sums to 788 s and wasm-jit's to 2216 s; the tail is
the Buildings EnergyPlus and Borefields models and the MSL machines
examples, which is exactly where the unfiltered files are widest.

Assisted-by: Claude Opus 5 (1M context)