[behdad]

BASE subsetting downgraded a 1.1 table to 1.0 when its variation-store offset was null. Instancing did the same after removing the store, and converted resolved format-3 coordinates to format 1. These changes preserved the coordinates but introduced unnecessary structural differences from HarfBuzz.

Preserve the source BASE version and its required null Offset32 field through ordinary subsetting and instancing. Keep format-3 coordinates after resolving their variation devices, following HarfBuzz's BASE subset path. A small writer wrapper retains the source version without changing the public owned BASE representation.

Validation: all 215 skera unit tests, three new BASE integration tests, existing layout precision tests, Clippy, formatting, and the CI typos checker pass. Three synthetic table references match fresh local HarfBuzz c82300aefb. BASE tables also match in 27 ordinary subset cases and 49 full/partial instancing cases across nine and six source fonts respectively, with no shaping differences.

Stacked on #2274.

Readers use read-fonts directly. write-fonts handles font assembly and owned table rebuilding; Skera retains its direct serializers and repacker. There is no separate generated Skera table tree.

[behdad]

Self-review completed at the prepared #2275 boundary:

• The source version determines the BASE header length. A version 1.1 table retains its null Offset32 field after the store disappears; a version 1.0 table does not acquire that field. I checked both direct subsetting and the instancing writer wrapper, including offset bases and delegation to owned-table validation.
• Keeping a resolved BaseCoord format 3 is intentional and agrees with HarfBuzz. Ordinary hint devices remain present, while fully pinned variation devices disappear after their deltas are folded.
• The small VersionedBase wrapper retains source metadata that the owned writer otherwise infers from store presence. It avoids adding another owned table model or generated code.

No new defect found in this slice. All three BASE integration tests pass at the prepared boundary, including horizontal/vertical axes, null stores, pins/ranges, and a subsequent subset. They also pass with all three references freshly regenerated using current HarfBuzz. Parent layout precision and ownership corrections remain included.

[behdad]

The reviewed parent fixes are now pushed. No new BASE defect was found: all three integration tests pass at this boundary with freshly regenerated HarfBuzz references, including the null 1.1 store field and retained coordinate formats.

[behdad]

Superseded by the rebuilt, consolidated instancing stack in #2416–#2423. MATH and name-table overrides remain separately covered by #2415 and #2411.