[morsel-ray]

Landed on dev as d4899d91cfbf4ee4ecaf7bbbd0f9ea480c517ff2 after independent exact-head approval5621634826 and eleven successful required checks. Complete checked/landed tree matched; no engine behavior change.

Change

Documentation portion of #948. ray_pmap_fn currently delegates directly to ray_map_fn; callback invocation is sequential even for large inputs or a larger -c setting.

• Correct both function-reference entries, math guide and analytics recommendation.
• Separate sequential map/pmap callbacks from eligible vector/query kernels inside them.
• Qualify architecture threshold wording: operator/shape, workers and dispatch safety matter;65,536 is common, not automatic parallelism for every query or lambda.
• Point independent task sweeps at existing multiple-instance guidance and per-process worker budgeting.

Five Markdown files only. No engine/API/pool change, parallel-lambda implementation, timing or memory-sharing promise. Feature/design work remains separate; no automatic issue-closing directive.

Validation

• mkdocs build --strict passes; actual rendered pmap-f-x and thread-pool anchors checked.
• Existing pmap filter 3/3 and map/pmap/vector examples pass using preserved ASan/UBSan binaries from reviewed e76dab11; relevant map/VM/test source blobs match accepted62d9. Not a fresh final-head engine build or full suite.
• Exact examples: (1 4 9) for both maps, [1 4 9 16 25] for vector multiply, (4950 19900 44850) for nested sum/til. Relevant whole diagnostic logs empty; binary hashes unchanged.
• Whitespace/docs-only delta checked. Reporter timings, native thread participation, process-memory comparison and actual parallel callback execution not independently tested.