perf(extract): coalesce archive blocks into large writes - #1681

Summary

Multi-file extraction performs one disk write for every small libarchive data block; this coalesces contiguous blocks into 8 MiB writes while preserving exact offsets and error behavior.

Why This Exists

archive_write_data_block() has meaningful per-call overhead, especially on removable media with write caching disabled. In device testing, batching reduced a representative extraction from approximately 26 minutes to 5.5 minutes.

Resolution

Add a small, independently tested BlockBatcher between the archive reader and archive_write_disk. Contiguous blocks are buffered, offset gaps flush immediately to preserve sparse files, and blocks at least as large as the buffer bypass the copy.

Reviewer Considerations

• The buffer is 8 MiB per active extraction thread.
• Sparse and backwards offsets force a flush rather than being merged.
• Pending data is flushed before archive_write_finish_entry().
• Write errors propagate unchanged and failed data is not retried implicitly.

Behavior Changes

Multi-file images are written using fewer, larger disk operations. Extracted bytes and offsets remain unchanged.

Implementation Summary

• Add a header-only contiguous block batcher.
• Integrate it into multi-file extraction.
• Add 11 focused tests covering batching, sparse offsets, boundaries, bypasses, and failures.
• Log source block size diagnostics for field validation.

Verification

• ./build/test/block_batcher_test: 122 assertions in 11 test cases passed.
• Full macOS build compiled downloadextractthread.cpp successfully; final universal-app link was blocked by the local arm64-only Homebrew Qt installation.
• git diff --check: passed.

Risk

Low to moderate. The buffering path is isolated and extensively tested; the primary tradeoff is an additional 8 MiB allocation during multi-file extraction.