Fix pipeline callbacks that fail under multiprocessing pickling

[Ulthran]

Motivation

• Tests were failing because callbacks (local functions / lambdas) passed to the pipelines were not pickleable when using process-based pools, causing AttributeError: Can't get local object during multiprocessing.
• The pipeline needs to support local callback functions while preserving chunked processing and counter aggregation.

Description

• Replaced process-based multiprocessing usage with ThreadPool by importing ThreadPool from multiprocessing.pool and using it for parallel execution in src/heyfastqlib/pipelines.py.
• Added a fast-path for single-threaded execution (threads == 1) that calls _filter_worker / _map_worker inline to avoid any pool overhead or pickling.
• Kept existing chunking via _chunk_reads and counter aggregation via _merge_counters so read/base accounting remains unchanged.

Testing

• Ran black . to format the code, which completed successfully.
• Ran pytest tests, which passed with all tests succeeding (36 passed).

---

Codex Task

[Copilot]

Pull request overview

This PR fixes multiprocessing pickling failures when local functions or lambdas are passed as callbacks to pipeline functions. The issue occurred because the previous implementation used process-based Pool from multiprocessing, which requires serializing (pickling) all arguments including callback functions.

Changes:

• Switched from process-based Pool to ThreadPool to avoid pickling requirements for callback functions
• Added single-threaded fast-path optimization when threads == 1 to bypass pool overhead
• Preserved existing chunking and counter aggregation behavior

---

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.