Avoid thread fanout for single-item parallel maps

[jwils]

Summary

• add a fast path so ElasticGraph::Support::Threading.parallel_map uses normal map for exactly one item
• keep multi-item enumerables on the existing thread-per-item path
• add a benchmark plus captured results for the single-item fanout case
• cover single-item arrays, single-entry hashes, multi-entry hashes, and enumerables without size

Context

This avoids creating a Ruby thread for single-client msearch fanout, which is common for deployments that route queries through one datastore cluster.

Benchmark

bundle exec ruby benchmarks/threading/parallel_map_single_item.rb

• single item array: after path 3.56M i/s vs before 24.3k i/s (146x faster)
• single entry hash: after path 3.23M i/s vs before 27.6k i/s (117x faster)
• Thread.new calls for 1,000 single-item maps: before 1000, after 0
• multi-item array: same-ish within benchmark error
• empty array: current implementation is slower in this microbenchmark (208.29ns vs 128.74ns) because it pays the one-item shape check, but neither path creates threads

Testing

• bundle exec rspec elasticgraph-support/spec/unit/elastic_graph/support/threading_spec.rb
• script/run_gem_specs elasticgraph-support
• script/lint elasticgraph-support/lib/elastic_graph/support/threading.rb elasticgraph-support/spec/unit/elastic_graph/support/threading_spec.rb benchmarks/threading/parallel_map_single_item.rb
• script/type_check
• bundle exec ruby benchmarks/threading/parallel_map_single_item.rb

Generated with Codex

[myronmarston]

👏 Great find! Left a few minor nits.