Fix ordered BatchtoolsParam iteration without reduction

[LeonidasZhak]

Summary

Allow bpiterate(..., reduce.in.order = TRUE) to return ordered results with BatchtoolsParam when no REDUCE function is supplied.

Thanks to maintainers

Thanks to the BiocParallel maintainers for maintaining the parallel backends and for the guidance in #286.

Issue or motivation

Fixes #286. SerialParam, MulticoreParam, and SnowParam accept this combination, but BatchtoolsParam rejected it before the shared reducer could collect the results.

Root cause

The BatchtoolsParam-specific bpiterate() method unconditionally stopped when REDUCE was missing and reduce.in.order was TRUE. The shared IterateReducer already supports ordered collection without reduction.

Change

Remove that extra rejection and add regression coverage for the multicore and socket Batchtools backends. The existing error for supplying init without REDUCE remains unchanged.

Tests

• Baseline check on current devel: reproduced the original REDUCE must be provided when 'reduce.in.order = TRUE' error.
• Focused BatchtoolsParam regression: passed for multicore and socket backends.
• Full BiocParallel RUnit suite: 110 test functions, 0 errors, 0 failures; 3 MPI/Rmpi tests deactivated because Rmpi is not installed locally.
• R CMD INSTALL --no-multiarch --with-keep.source .: passed.
• R CMD build .: passed, including vignette creation.
• _R_CHECK_FORCE_SUGGESTS_=true R CMD check --no-manual BiocParallel_1.47.0.tar.gz: passed with 1 existing WARNING and 2 existing NOTEs.
• I also tried R CMD check --as-cran --no-manual; it fails on existing examples/tests because CRAN's _R_CHECK_LIMIT_CORES_ cap conflicts with existing MulticoreParam(3)/larger-worker examples, outside this PR's diff.

Scope

This changes only the BatchtoolsParam bpiterate() guard and its unit tests. It does not change reduction behavior for supplied REDUCE functions, init validation, other backends, batchtools scheduling, documentation, or vignettes.