[Docs] Document filetype-aware input-extension defaults in deduplication workflows #2127

Context

Child of #2118. #2045 changed exact, fuzzy, semantic, and removal workflows so omitted input_file_extensions derives from input_filetype instead of generic partitioner defaults. This is useful behavior but is absent from the relevant workflow references.

Requirements

• Document the default extension set for every supported input_filetype.
• Explain the distinction between omitted, explicit list, and empty/invalid values.
• Cover exact, fuzzy, semantic, and duplicate-removal workflows consistently.
• Include the JSONL example from the PR and at least one Parquet example.
• Verify the existing fuzzy input_blocksize clarification from docs: clarify fuzzy dedup input blocksize #2096 remains correct and nearby terminology is consistent.

Acceptance criteria

• Relevant dedup workflow pages show the derived defaults
• Examples demonstrate explicit override behavior
• Unsupported combinations and recursive discovery behavior are clear
• Cross-workflow terminology is consistent
• Fern checks pass

Related PRs: #2045, #2096