Fix undetected buffered output-write failures

[thierrygosselin]

Summary

This PR makes the buffered Writer fail explicitly when output cannot be written completely.

Previously, writeInternal() accepted any positive fwrite() result as success, callers could ignore a failed write, and final stream errors were unchecked. This could leave truncated output while fastp returned exit status 0.

Changes

• Check complete writes and stream errors.
• Fail explicitly on compression-buffer allocation or compression failure.
• Check fclose() errors.
• Check final flushing of stdout without closing it.
• Include the output destination in error messages.

The separate multithreaded gzip pwrite implementation is unchanged.

Regression tests

Added scripts/test_buffered_output_errors.py, covering plain FASTQ, single-thread gzip, stdout, and split output.

The tests use synthetic reads and a process-specific file-size limit to simulate write failures without filling a disk. Reports are directed to /dev/null to isolate FASTQ output failures.

Run with:

python3 scripts/test_buffered_output_errors.py ./fastp

Tested locally on macOS ARM64:

• Unmodified Conda fastp 1.3.6: all four normal-output cases pass; all four write-error cases incorrectly exit successfully.
• Locally compiled patched source: all eight cases pass.
• Successful outputs preserve the synthetic reads exactly.

Linux execution and isolated allocation/compression-failure injection have not been tested.