Use klauspost/compress for tar and zip decompression

[andrew]

The module already depends on klauspost/compress for zstd; its gzip subpackage is a drop-in replacement for stdlib and aliases ErrChecksum/ErrHeader to the stdlib sentinels, so no callers or existing tests change.

Interleaved benchstat, arm64 darwin, n=10:

                              │   before    │             after              │
                              │   sec/op    │   sec/op     vs base           │
OpenBytesTarGz                  552.8µ ± 6%   437.1µ ± 4%  -20.93% (p=0.000)
TarBrowse/test.tar.gz/0         12.48µ ± 5%   11.68µ ± 3%   -6.40% (p=0.005)
TarBrowse/test.tar.gz/512       17.93µ ± 4%   16.19µ ± 4%   -9.72% (p=0.000)
TarBrowse/test.tar.gz/16384     37.18µ ± 4%   33.97µ ± 3%   -8.63% (p=0.000)
TarBrowse/test.tar.gz/65536     90.14µ ± 3%   84.65µ ± 5%   -6.09% (p=0.000)
TarBrowse/test.tar.gz/1048576   1.157m ± 4%   1.068m ± 3%   -7.71% (p=0.000)
geomean                         88.47µ        79.56µ       -10.07%

An isolated decoder A/B on text-like data (25% compression ratio, closer to real source archives than the random-payload fixtures) shows 1.44–1.88×. B/op drops 0.03–9.3% across the .gz cases; allocs/op gains a fixed +2 per open from the reader setup.

Stripped binary size for a minimal consumer grows by ~84 KB (+3.1%) since archive/zip keeps stdlib compress/flate linked, so klauspost's flate is additive rather than a replacement.

Adds a multistream-gzip test (Alpine .apk shape) and a corrupt-header test; both pass against stdlib and klauspost.

For zip, reader.RegisterDecompressor(zip.Deflate, ...) swaps in klauspost's flate on the existing archive/zip reader. That bypasses archive/zip's internal flate pool, and klauspost's Reset wraps any input lacking io.ByteReader in a fresh bufio.Reader (archive/zip passes an io.SectionReader), so zip.go pools a *bufio.Reader alongside the decoder and hands that to Reset. Zero binary cost since klauspost's flate is already linked via gzip. Adds BenchmarkZipExtract (64×16 KiB deflated members, ~34% ratio):

             │   before    │             after              │
             │   sec/op    │   sec/op     vs base           │
ZipExtract     3.785m ± 3%   2.350m ± 3%  -37.91% (p=0.000)
             │    B/op     │    B/op      vs base           │
ZipExtract    50.96Ki ± 2%  48.33Ki ± 1%   -5.14% (p=0.000)
             │  allocs/op  │  allocs/op   vs base           │
ZipExtract      826.0 ± 0%    722.0 ± 0%  -12.59% (p=0.000)

conda.go is left alone: its zip members are Store, so registering a deflate decompressor there does nothing.

[Copilot]

🟢 Approval recommended

No unresolved blocking issues were identified.

Pull request overview

This PR switches TAR gzip decompression to Klauspost’s compatible implementation for improved performance while preserving behavior.

Changes:

• Use klauspost/compress/gzip for TAR decompression.
• Add multistream and corrupt-header tests.

File summaries

File	Description
tar.go	Uses Klauspost gzip for TAR decompression.
tar_read_test.go	Adds multistream and header-error coverage.

Review details

• Files reviewed: 2/2 changed files
• Comments generated: 0
• Review effort level: Lite

---

💡 Add a code-review agent skill or configure MCP servers for context-aware, tailored reviews. Learn more in the docs.

[Copilot]

🟢 Approval recommended

No unresolved blocking issues were identified, and all reviewed changes preserve behavior while improving decompression performance.

Review details

• Files reviewed: 4/4 changed files
• Comments generated: 0 new
• Review effort level: Lite