fix(physical-plan): CoalescePartitionsExec panic on wasm32-unknown-unknown

[kentkwu]

Which issue does this PR close?

• Closes UNION ALL panics on wasm32-unknown-unknown #24886

Rationale for this change

Multi-partition queries panic on wasm32-unknown-unknown when they funnel through CoalescePartitionsExec — most commonly UNION ALL, but any plan whose intermediate results have more than one partition (e.g. a multi-file external table scan) trips this. The operator's multi-input branch spawns via JoinSet::spawn, which requires a tokio reactor that isn't installed under wasm-bindgen-futures:

panicked at datafusion/common-runtime/src/join_set.rs:69:20:
there is no reactor running, must be called from the context of a Tokio 1.x runtime

This is a follow-up to #24275, which fixed the same class of panic in collect_partitioned.

What changes are included in this PR?

This PR adds a single-threaded path to CoalescePartitionsExec::execute: when the session's target_partitions equals 1, the operator drains its input partitions sequentially instead of spawning a task per partition.

Behavior when target_partitions > 1 is unchanged. Since the native default is num_cpus::get(), only callers who have explicitly opted out of parallelism take the new path.

Are these changes tested?

Yes.

• Adds a test_union_all regression in datafusion/wasmtest that runs under both wasm-pack test --chrome and tokio::test.
• Confirmed the reproducer panics on unpatched main and passes with the patch applied.
• cargo fmt --check and cargo clippy -- -D warnings both run clean.

Are there any user-facing changes?

There are no API changes. On wasm32-unknown-unknown, multi-partition queries no longer panic when the session has target_partitions = 1. Native callers running with the default target_partitions > 1 are unaffected.

[yinli-systems]

I reproduced the browser panic and reviewed the scheduling and regression-test behavior. I don't think this should merge as-is because the fallback changes native scheduling and can stop polling exchange-backed partitions concurrently; the test also relies on an output order the operator does not guarantee. The inline comments include a wasm-only cooperative alternative and the validation I ran.

[codecov-commenter]

Codecov Report

✅ All modified and coverable lines are covered by tests.
✅ Project coverage is 81.62%. Comparing base (d02bd78) to head (cbb7717).

Additional details and impacted files

@@            Coverage Diff             @@
##             main   #24890      +/-   ##
==========================================
- Coverage   81.63%   81.62%   -0.01%     
==========================================
  Files        1123     1123              
  Lines      409546   409558      +12     
  Branches   409546   409558      +12     
==========================================
+ Hits       334319   334321       +2     
- Misses      55591    55601      +10     
  Partials    19636    19636              

☔ View full report in Codecov by Harness.
📢 Have feedback on the report? Share it here.

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.
• 📦 JS Bundle Analysis: Save yourself from yourself by tracking and limiting bundle sizes in JS merges.

[kentkwu]

Thanks for the review @yinli-systems. I agree with the recommended changes and applied them

[yinli-systems]

Thanks for addressing both points. The wasm-only select_all path preserves native scheduling and cooperative polling across input partitions, and the sorted assertion now matches the operator contract. I reviewed the updated diff and the full CI suite is green.