Preserve raw XBRL and tighten validation

[ThomasMarcelis]

This PR makes the parsed output keep more of the original XBRL instance document, and makes Validate() report structural problems more explicitly.

Main changes:

• RawXBRL / XBRL now preserve the root XML name and attributes, raw context/unit slices, and raw reference elements such as schemaRef, linkbaseRef, roleRef, arcroleRef, and footnoteLink.
• Segment elements now preserve nested XML through InnerXML.
• Context scenario elements are preserved as raw XML, but still rejected by Validate() because scenario support is not implemented.
• Validate() now checks contexts, units, duplicate IDs, period shape, entity identifiers, fact structure, broken fact references, and unsupported top-level item / tuple elements.
• Fact.Validate() now returns specific errors for malformed numeric, non-numeric, nil, and fraction facts.
• NumericValue() now returns errors for malformed numeric facts instead of panicking on missing fields.
• Added Parse, ParseReader, and Decode helpers around the existing encoding/xml flow.

This is mostly additive API surface, but validation is intentionally stricter: some malformed documents that previously parsed without a useful validation error may now fail Validate().

Tests cover focused XML snippets, the existing AAPL and EDGR fixtures, examples, and parse/validate fuzzing.

Checked with go test ./....