Fix list-valued generate/analyse on parallel workers (future + socket)

[emstruong]

Fix list-valued generate/analyse on parallel workers (future + socket)

Claude wrote this, I don't have the understanding of how parallel processing in R works to review this, sorry. But I thought perhaps it might save you time by having Claude's proposed changes + tests.

Closes #90.

What

Make list-valued generate/analyse work on every back-end. Two gaps caused fatal
terminations when the function lists (stashed in SimDesign:::.SIMDENV) were not
delivered to parallel workers:

1. parallel = "future" — neither GENERATE_FUNCTIONS nor ANALYSE_FUNCTIONS
   reached the workers (future's automatic global detection does not export
   package-internal objects), so the run aborted with
   object 'ANALYSE_FUNCTIONS' not found (or 'GENERATE_FUNCTIONS').
2. parallel = TRUE (socket/MPI) — a list-valued generate aborted with
   object 'GENERATE_FUNCTIONS' not found, because clusterExport() shipped
   ANALYSE_FUNCTIONS but never GENERATE_FUNCTIONS.

Serial (parallel = FALSE) was always fine.

Why

When generate/analyse are lists, runSimulation() substitutes the internal
combined_Generate()/combined_Analyses() dispatchers and stores the original
function lists in the package-internal .SIMDENV. Workers load a fresh copy of the
package whose .SIMDENV is empty, so the lists must be sent to them explicitly. One
list was missing from the socket export, and the future path did not export them at
all.

How

• R/runSimulation.R: add a local GENERATE_FUNCTIONS (mirroring ANALYSE_FUNCTIONS)
  and a guarded clusterExport(cl, "GENERATE_FUNCTIONS").
• R/analysis.R: in the future branch of Analysis(), wrap the worker function so
  every future restores the captured function lists into the worker's .SIMDENV via
  getFromNamespace() before the replication runs — correct for any future plan.

Tests

tests/tests/test-01-core.R now exercises list-valued generate and list-valued
analyse across serial, socket (parallel = TRUE) and future (multisession)
back-ends, asserting no FATAL_TERMINATION.

Verification

Built and installed the patched package; both reprexes return FATAL_TERMINATION = NULL and correct output columns across 3/3 trials on each affected back-end. The same
reprexes fail deterministically (3/3) on the released package.

Note for reviewers: reproduce with the installed package, not devtools::load_all()
— load_all() masks the future failure because the dev namespace makes future
serialise .SIMDENV as an ordinary global.

[philchalmers]

It's awkward to do the redefinition of mainsim, but I don't see away around this since future doesn't expose the cluster info. Could you add this information to the NEWS.md file?

[emstruong]

Hi @philchalmers , I asked Claude to give it a shot.

[philchalmers]

Good enough for me. I'll make any edits after the merge. Thanks!