Expose stream, seek, and span archive input modes

[Hawkynt]

What changed

• Added explicit *Streaming, *Seekable, and *Span read/list/extract entry points to IArchiveFormatOperations.
• Kept the existing Stream methods as the compatibility/native descriptor hooks, so every existing archive and pseudo-archive gains all three source modes without duplicating adapter code.
• Added a large-file-safe bridge for legacy seek-only readers: forward-only input is spooled to a delete-on-close temporary file rather than buffered into a managed array.
• Added ownership wrapping so a spool or span-backed compatibility buffer stays alive for the lifetime of a returned entry stream.
• Added extension methods so default interface members are also ergonomic when descriptors are referenced by their concrete type.
• Added registry-level NUnit tests using an intentionally seek-only fake descriptor to prove the universal fallbacks rather than one format-specific implementation.

Design decisions

Libarchive was used as architectural reference, not as implementation source. Its archive-read API treats forward reads as the primary callback, exposes seek separately, and supports memory-backed input. CompressionWorkbench now exposes the same three input shapes at the common archive/pseudo-archive boundary while preserving existing descriptor implementations.

The compatibility streaming path does not pretend a legacy random-access parser is genuinely one-pass: it spools only when the caller supplies a non-seekable stream. Formats with true sequential parsers can override ListStreaming, ExtractStreaming, and OpenEntryStreaming. Native random-access parsers can override the *Seekable path, and native span parsers can override ListSpan, ExtractSpan, and OpenEntrySpan to avoid the compatibility copy.

The span API uses explicit *Span names instead of overloads. An initial overload design made existing XML references to IArchiveFormatOperations.List ambiguous throughout the repository; explicit mode names preserve source/documentation compatibility and make the three source capabilities obvious at call sites.

No new dependencies were introduced. No libarchive source code was copied or translated.

Reference material

• libarchive archive.h read/skip/seek callback contract
• libarchive project documentation describing callback-driven stream and memory access
• libarchive BSD license reviewed for provenance; only public API behavior/architecture was consulted

Validation

• Repository README check: green on current CI run.
• Ubuntu Release build: green on current CI run.
• Windows build and the remaining test/interop/package checks are still running on the current head.
• Earlier CI iterations caught and drove fixes for malformed XML cref syntax and the ambiguous span-overload design.