Performance: major runtime optimizations for parallel simulation, spatial joins, raster covariates, and result aggregation PR Description

[hambrecht]

Summary
This PR introduces four major performance changes to the simulation pipeline, each isolated in its own commit.
The goal is to reduce wall-clock time, improve CPU utilization, and lower memory/serialization overhead for large runs.

What Changed

1. Parallel execution redesign

• Switched to a worker-state model so workers initialize shared state once.
• Replaced static task distribution with load-balanced scheduling.
• Reduced repeated argument serialization overhead in parallel loops.
• Key file: run.simulation.R

2. Spatial distance computation redesign

• Replaced per-transect intersection loops with indexed spatial joins.
• Switched to vectorized distance computation paths.
• Removed iterative reduce/rbind assembly bottlenecks.
• Key files:
  • calc.perp.dists.R
  • calc.rad.dists.R

3. Raster covariate caching

• Added process-local caching for cov.surface rasters and value vectors.
• Avoids reloading/re-extracting full raster values every repetition.
• Improves repeated scale-parameter calculation performance when raster covariates are enabled.
• Key file: calculate.scale.param.R

4. Result transport and aggregation redesign

• Changed per-repetition return shape to compact rep-level payloads.
• Added rep-level extract/apply helpers for lower transfer and copy costs.
• Updated serial and parallel aggregation paths to consume compact payloads.
• Maintains backward-compatible accumulation path for legacy result shape.
• Key files:
  • single.sim.loop.R
  • accumulate.PP.results.R
  • run.simulation.R

Expected Runtime Impact

1. Better core utilization in heterogeneous replicate workloads due to load balancing.
2. Lower parallel overhead from reduced repeated serialization of large objects.
3. Faster geometry-heavy runs due to indexed spatial joins and vectorized distance calculations.
4. Faster raster-enabled runs due to worker-local cache reuse.
5. Reduced memory pressure and lower inter-process transfer volume during result aggregation.

Validation

1. Targeted raster feature regression test passed fully: test-check_RasterCovariate.R
2. Result: PASS 11, FAIL 0, WARN 0, SKIP 0

Commit Structure

1. perf(parallel): switch to worker-state load-balanced execution
2. perf(spatial): vectorize detection-distance joins with sf indexes
3. perf(raster): add worker-local cov.surface cache
4. perf(results): stream per-repetition payloads instead of full result objects