Module parallel 

Rayon-backed parallel split + recombine, plus a branch-level ParallelExecutor for independent plan fragments.

Two parallelism shapes live here:

• run_parallel — chunked Vec-in / Vec-out. Used by parallel- aware operators (large hash joins, blocking sorts, hash aggregates) to fan out per-partition work.
• ParallelExecutor — runs N independent worker closures concurrently and returns their results in input order, allowing the operator-tree driver can fork independent branches (Intersect / Union / BayesianEvidenceFusion / RobustPositiveEvidencePool / ProbBoolFusion children, deep-fusion SignalLayer signals) without serialising them.

Structs

ParallelExecutor

Branch-level parallel executor.

Constants

DEFAULT_PARALLEL_WORKERS

Default thread-pool size. Setting the runtime value to 0 disables parallel execution.

MIN_PARALLEL_BRANCHES

Minimum number of branches before parallel dispatch kicks in. Below this, sequential execution wins on overhead.

Functions

run_parallel

Split input into num_partitions chunks, run worker over each chunk in parallel, and concatenate the results in the same order the chunks were emitted in.