35.8.0

Changed

• Removed the primary-key-only limitation from IterateById: reusable projections can now include stored value fields,
  deserialized using the row's schema version while preserving [NotStored] context.
• Added FastIterateNoCursor for callbacks that do not access the cursor, with deferred cursor/index updates
  and optimized short key suffix copies. Existing FastIterate continues to expose the current cursor during callbacks.
  Primary-key-only IterateById scans use the new API.
• Further optimized FastIterateNoCursor with padded short-suffix word copies, fewer bounds checks, and
  a single per-key iteration counter.
• Reduced work when scanning prefixed BTree leaves by combining scratch-buffer capacity checks, copying short
  prefixes with a single word, reusing adjacent suffix offsets, and simplifying branch cursor-position updates.

Added

• Added IterateById relation methods that hydrate a projection into one caller-provided instance for every matching
  primary-key row, allowing allocation-free callbacks and preserving [NotStored] context on the reused instance.