[ViktorBystrov72]

net.Buffers.WriteTo modifies its slice entries. When those entries share the backing array of Buffer.bufs, the pool-return loop receives nil or shortened slices instead of the original chunks.

Copy the completed-chunk descriptors before writing. The chunk data and vectored-write path stay the same, and chunks are still offered back to the pool only after WriteTo returns. Empty and single-chunk buffers keep the existing path.

Fixes #359.

Local microbenchmark, Go 1.23.0 on darwin/arm64, six samples, 16 KiB written to io.Discard:

Metric	Before	After
Bytes allocated/op	16,699	946
Allocations/op	10	11

The measured case trades one small descriptor allocation for reuse of the larger chunks. This is a local result, not a universal performance claim; allocation counts for the empty and 128-byte cases were unchanged.

Validation:

• A diagnostic probe showed original chunk capacities becoming zero before the pool-return loop; the fix preserves them.
• Partial-write/error tests check byte counts, written prefixes, reset, and reuse. They do not depend on sync.Pool retention across GC.
• Full make (generation, tests, benchmarks, and golint), go test -race -count=1 ./..., and go vet ./... passed.