Which solver integrates a model is a simulation flag, so the four jobs
translated and compiled the same 19500 models to run them four times.
`--solver` builds each model once and runs its executable once per
solver, the same bargain `--fmisimulator` and `--wasmjitrunner` strike:

    ./test.py --branch=master --solver=default,cvode,gbode,ida ...

`default` is the solver the model asks for and every other name is what
the run is given as `-s`, from configs/solvers.json, so adding a solver
is an entry there and no change to any script. Each fills the table its
entry names - master, cvode, gbode, ida - so no report and no history
has to move, and the four Jenkins stages become one on master's node.
The old `-d=newInst,-frontEndUnitCheck` is gone with them: newInst
defaults to true and frontEndUnitCheck to false, so master's build is
what it was.

The branches are not filled at the same rate - master twice a day,
cvode and gbode weekly, ida now and then - so a library is now only run
for the runners that are missing it, rather than being skipped whenever
the first one has it. A weekly `default,cvode` therefore builds a
library master already has and simulates it with CVODE alone. The FMI
and wasm-jit jobs get the same rule.

The first runner of a library also writes its result and difference
files under the plain names again. It has been the branch's own
directory that publishes them unsuffixed since the FMI jobs were
merged, so a job with more than one runner published the first one's
diff pages under names its report did not link.

Checked on the sanity check configuration: four solvers fill their four
tables from one build, a re-run skips everything, and dropping the
cvode results and running again rebuilds the libraries and simulates
them with CVODE alone, leaving the other three tables untouched.

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