Fix duplicated, misordered and truncated Word/Powerpoint snapshots

[SimonCropp]

RegisterFileConverter and RegisterStreamConverter(ext) were both pointed at the same Convert* method. Verify runs the extension stream converter over the package target a file converter returns, so verifying a typed document (WordprocessingDocument, SpreadsheetDocument, PresentationDocument) ran the whole extraction twice and wrote the info, the text and the rendered pages to two snapshot files each. A file converter now just hands over the cloned package and lets the extension converter own the extraction, so the typed and stream paths produce byte-identical output.

Three further bugs in the Word converter:

• The document text was carried in WordInfo.Text and emitted as a txt target, putting it in two snapshots. It is now only the target. Same for PowerpointInfo.Text.
• Document order was lost. The walk collected every paragraph then every table, and within a run every text then every tab then every break. It is now a single in-order walk.
• Content controls were dropped, so all text inside a w:sdt vanished from the snapshot. SdtBlock, SdtRun, SdtRow and SdtCell are now transparent, matching what Word displays. Hyperlink runs were dropped too.

Adds Word tests for ordering, inline/block/in-cell content controls, hyperlinks and interleaved text and tabs. There was no w:sdt coverage, and sample.docx is two bare paragraphs, which is why none of this surfaced.

Snapshot churn across 15 tests is mostly filename shift rather than content: the duplicate #00/#1 pairs collapse to a single unsuffixed file.